Skeptical Linguistic Essays

Skeptical Linguistic Essays Paul M. Postal OXFORD UNIVERSITY PRESS Skeptical Linguistic Essays This page intention...

Author: Paul M. Postal

42 downloads 547 Views 1MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Skeptical Linguistic Essays

Paul M. Postal

OXFORD UNIVERSITY PRESS


This page intentionally left blank


Paul M. Postal

1 2004

3

Oxford New York Auckland Bangkok Buenos Aires Cape Town Chennai Dar es Salaam Delhi Hong Kong Istanbul Karachi Kolkata Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi São Paulo Shanghai Taipei Tokyo Toronto

Copyright © 2004 by Oxford University Press, Inc. Published by Oxford University Press, Inc., 198 Madison Avenue, New York, New York 10016 www.oup.com Oxford is a registered trademark of Oxford University Press All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press. Library of Congress Cataloging-in-Publication Data Postal, Paul Martin, 1936– Skeptical linguistic essays / Paul M. Postal. p. cm. Includes bibliographical references and index. ISBN 0-19-516672-8; ISBN 0-19-516671-X (pbk.) 1. Grammar, Comparative and general. 2. English language—Syntax. Noam. 4. Linguistic analysis (Linguistics) I. Title. P151.P655 2003 415—dc21 2003040455

2 4 6 8 9 7 5 3 1 Printed in the United States of America on acid-free paper

3. Chomsky,

CONTENTS

Introduction

3 PART I. STUDIES IN LINGUISTICS

1. 2. 3. 4. 5. 6.

A Paradox in English Syntax 15 A Putatively Banned Type of Raising 83 A New Raising Mystery 109 Chromaticity: An Overlooked English Grammatical Category Distinction 138 The Structure of One Type of American English Vulgar Minimizer 159 The Openness of Natural Languages 173 PART II. STUDIES OF JUNK LINGUISTICS

7. 8. 9. 10. 11. 12.

13. 14. Notes

Junk Syntax 1: A Supposed Account of Strong Crossover Effects 205 Junk Syntax 2: “There Remain a Few As Yet Unexplained Exceptions” 233 Junk Ethics 1: Advances in Linguistic Rhetoric 286 Junk Refereeing: Our Tax Dollars at Work 293 Junk Ethics 2: The Most Irresponsible Passage 296 Junk Reasoning 1: If It Doesn’t Follow Automatically, Then It’s Pretty Much Got to at Least Virtually Follow, and If Not, Don’t Worry; It Is Still Unquestionably Natural 306 Junk Reasoning 2: ‘(Virtually) Conceptually Necessary’ 323 Junk Linguistics: The Bottom Line 337 339

References Index

405

385




INTRODUC TION

T

his collection of essays is concerned with syntactic questions, with certain general features of grammatical theory related to syntax, here and there with semantic issues, and quite a bit with questions of appropriate standards in pursuing research in the previously mentioned domains. It has almost nothing to say about phonology. The immediately following remarks are to be interpreted against the background of this restricted understanding of what ‘linguistic’ is here intended to denote. A prospective reader might ask “what is a skeptical linguistic essay?” I would answer that it is one based on a deep and long-standing view that much nonetheless prestigious current linguistics has in fact made very restricted descriptive and explanatory progress and, in some areas where great things have been claimed, no real substantive progress at all. This lack of true progress holds, I maintain, despite the fact that the literature of modern linguistics and its ancestors is replete with boasts of at least grand-sounding accomplishments; see in particular the discussion of (2) of chapter 8. Consider (1) for example: (1)

Epstein and Hornstein (1999: xi) “GB’s very success, however, dramatically alters the methodological landscape. It has spawned a consensus that principles-and-parameters accounts may well answer Plato’s problem in the domain of language. This general and surely tentative consensus allows the other sorts of measures of success to guide minimalist theory construction and (self-)evaluation. In effect, given a principles-and-parameters setting, simplicity, elegance, parsimony, and naturalness move from background to foreground concerns in the con3

4

INTRODUCTION

struction and evaluation of analyses. To put matters more tendentiously than is warranted: given that principles-and-parameters models ‘solve’ Plato’s problem, the paramount research issue becomes which of the conceivable principles-and-parameters models is best, and this issue is (in part) addressed using conventional (not uniquely linguistic) criteria of theory evaluation.”

Various hedges notwithstanding, the authors of this passage clearly communicate the view that the government binding (GB) framework has been a deep success and has even in some sense solved the problem of natural language (hereafter: NL) acquisition (‘Plato’s problem’). This is in my view so far from reality as to be little more than a dream.1 The claim that GB has been a success is not backed up, and I claim could not be, by any citation of substantive factual results for the grammar of English, for example.2 Chapters 2, 3, 5, 7, and 8 consider in fact cases where some factual success in these terms seems to have been proclaimed and I suggest that they show that the claims are groundless. Note that passages like (1) make no attempt to consider criticisms of the favored view nor do they deal with arguments, many of considerable detail and depth, that other, competing views of NL syntax are far superior to the GB view. Work in lexical functional grammar (LFG) and head-driven phrase structure (HPSG), categorial grammar, and so on, is unmentioned. In short, such passages partake more strongly of the character of factually empty propaganda than of serious scholarship. Beyond any substantive questions, many of the essays that follow thus express even more serious skepticism about the standards that seem to govern the way much linguistic research is evaluated, justified, and promoted. Lack of actual progress is unfortunate but at least remediable. But inadequate standards can undermine the entire enterprise of linguistic inquiry by even eating away at a clear conception of what an actual result might be. One non-negligible aspect of this damaging development is seen in remarks like (2): (2)

Chomsky and Lasnik (1995: 28) “The shifts in focus over the years alter the task of inquiry considerably and yield different conceptions of what constitutes a ‘real result’ in the study of language. Suppose we have some collection of phenomena in a particular language. In the early stages of generative grammar, the task was to find a rule system of the permitted form from which these phenomena (and infinitely many others) could be derived. That is a harder task than the ones posed in pregenerative grammar, but not an impossible one: there are many potential rule systems, and it is often possible to devise one that will more or less work— though the problem of explanatory adequacy at once arises, as noted. “But this achievement, however difficult, does not count as a real result if we adopt the P&P approach as a goal. Rather, it merely sets the problem. The task is now to show how the phenomena derived by the rule system can be deduced from the invariant principles of UG with parameters set in one of the permissible ways. This is a far harder and more challenging task. It is an important fact that the problem can now be posed realistically, and solved in interesting ways in some range of cases, with failures that are also interesting insofar as they point the way to better solutions. The departure from the long and rich tradition of linguistic inquiry is much sharper and

INTRODUCTION

5

more radical than in early generative grammar, with problems that are quite new and prospects that appear promising.”

The notion to be critically commented on here is that in the linguistics advocated by the authors, constructing a grammar of a NL that gets the facts right putatively “does not count as a real result.” Rather, supposedly, a genuine result is only achieved when a successful rule system is deducible from universal grammar. While such remarks sound impressively deep, their implications at this stage threaten an elimination of the constraints on linguistic theorizing by precluding a serious factual basis for it. How much can anyone be expected to labor to actually try to account for the facts of actual NLs, internal to a point of view that proclaims at the start that success in such an endeavor is not even a result? But, as stressed long ago (Chomsky, 1957: 11), proper understanding of NL universals depends on successful descriptions: “More generally, linguists have been concerned with the problem of determining the fudamental underlying properties of successful grammars.” The thinly disguised contempt for getting the facts right manifest in a passage like (2) emerges particularly clearly when one is told that “failures are also interesting” and “point the way to better solutions.”3 Such claims can only be seen as selfserving defensive attempts to make descriptive failure seem acceptable. Some failures (of some kind) might be interesting in some way, ‘interesting’ being monumentally subjective, and some might point the way to a better solution. Others might just be hopelessly wrong, misguided, contradictory, ludicrous, and hence incapable of pointing to anything positive whatever. But since supposedly even descriptive failures are valuable, then, assuming descriptive successes are of at least some value, too, passage (2) would appear to embody the overall idea that anything done in this framework is progress. While this view would be marvelously convenient for its supporters, it is evidently illusory! Implicit in (2) is the fantastic and unsupported notion that descriptive success is not really that hard and so not of much importance. Were such a claim true, it would be straightforward to show that it is. The authors would merely need to document the existing proper descriptions (since supposedly “it is often possible to devise one that will more or less work”). For of course they could not know that proper descriptions are easily obtained unless they had such available. Given the now very long period of work in the advocated terms, successful descriptions should be abundant. Since their current position is explicitly a development of decades of effort by hundreds of people, with English the favored domain of study, correctness of their point of view would, one would think, almost inevitably have yielded some (many?) concrete descriptive results, at least for English. But they offer nothing of the sort and it is not that common to see such things even attempted in recent years. One is a very long way indeed from concern with realizing the earlier goal (Chomsky, 1957: 13) of a grammar that would describe all the well-formed sequences and none of the ill-formed ones. Moreover, one need only look at various syntactic areas where the authors themselves have worked, passives, expletives, wh constructions, object raising constructions, parasitic gaps, control phenomena, crossover phenomena, ellipsis, negation, and so on to see that there is no subdomain, even in English, where a genuinely successful description can be said

6

INTRODUCTION

to have been achieved. This is argued in considerable detail for crossover phenomena and passives in chapters 7 and 8. In such a context, talk about ‘real results’ that go beyond successful grammars is largely make-believe. Challenged here, though, is not the idea that there are higher level results to be sought that would involve a reduction of aspects of successful descriptions to universal principles. Rather, at issue is the merit of skepticism about the ill-supported assumption that successful grammatical descriptions are relatively straightforwardly come by, thereby rendering notions of deducing such descriptions from universal principles currently meaningful. Perhaps the most negative aspect of claims that descriptive success is not a real result and that it is appropriate at this stage of linguistics to concentrate on higher order explanatory goals is that it undermines understanding of (i) the need for truly intensive factual study and (ii) the lack so far even in well-studied NLs like English of serious descriptive understanding. If one had achieved an adequate description of English passive phenomena, for example, in the almost half a century of work in the terms the authors favor, it would make sense to inquire into the general principles from which such a description followed. But as touched on in chapter 8, the actual descriptions offered in these frameworks so far are not only not successful but also so bad as to hardly merit being taken seriously. Just so with issues of the strong crossover phenomenon, as treated in chapter 7. Many claims of virtues for such descriptions are achieved only via a disregard for serious standards. But if there are no actual defensible descriptions, talk of deducing such from universal grammar or the like is evidently more like make-believe than serious inquiry. A last point: Dismissiveness with respect to the difficulty and importance of reaching adequate descriptions can only undermine the proper valuation of real work that actually advances descriptive adequacy. I am thinking here, for example, specifically of work on negative polarity, which I have tried to educate myself about over the last half-dozen years or so. Here it is uncontrovertible that really serious insights have been gained, specifically in the rich descriptive and theoretical studies partially and inadequately referenced by Atlas (1996, 1997, 2000), Beghelli (1994), Fauconnier (1975a, 1975b, 1976, 1979), Giannakidou (1998, 1999, 2001), Heim (1984), Hoeksema (1983, 1986, 1996, 2000), Horn (1995, 1996a, 1996b, 2001), Horn and Kato (2000), Horn and Lee (1995), Israel (1996, 1998), Kempson (1985), Krifka (1990, 1994, 1995), Ladusaw (1980, 1996a, 1996b), Linebarger (1981, 1987, 1991), van der Wouden (1994, 1997), and Zwarts (1995, 1996, 1999). But enormous problems remain and no one who even samples the literature on this topic can, I believe, be long deluded into shallow assumptions about the ease of finding viable descriptions. Anyone who dares to speak in such terms should, I suggest, be faced with the simple requirement of presenting a characterization of the distribution of, for example, English negative polarity any that, as in (2), “will more or less work.” The fourteen essays found here, only one of which was previously published, fall into two categories. Part I contains six chapters that purport to be positive contributions to the study of English syntax, with implications for general syntactic theory. While the point is not made explicitly in any of them, I do not think it possible to reflect on their content without seeing how far one remains in the year 2003 from serious understanding of English grammar.

INTRODUCTION

7

Chapter 1 deals with the vexed question of English so-called locative inversion structures like (3): (3)

a. Under the rock was lying a large snake. b. In the sphere could be seen several space aliens. c. At that time were invented new forms of defensive rhetoric.

This chapter, mostly written in 1994–1996, arose as a response to Bresnan (1994), which argued that the initial PPs in (3a, b) are subjects. Chapter 1 develops a range of evidence that they are not and that such clauses have invisible expletive subjects. An interesting consequence of these conclusions, if true, is that the evidence that Bresnan (1994) took to show the subjecthood of the relevant PPs actually argues that these phrases fall under the category of noncanonical DP behavior, known principally in the case of ‘unexpected’ or ‘quirky’ case marking, for example, dative for subjects. The discussion reveals that it is all too easy to draw an erroneous conclusion on the basis of a few (real) arguments that favors a position one finds pleasing because that pleasing character may deter one from seeking evidence against that position. In effect, one is always balancing the pros and cons of various proposals and one must beware of prematurely concluding P on the basis of evidence for it if one has not systematically considered the question of available evidence against it. Chapter 2 addresses a notable convergence of views among the GB, LFG, and HPSG frameworks. For quite distinct theoretical reasons, all three of these positions developed during the 1980s came to embody the claim that there can be no phenomenon in NL syntax that could be informally characterized as “raising to complement/ object of P.” This might, for instance, be the proper analysis of cases like (4), as in fact suggested in Postal (1974: 363, note 7): (4)

Dora was depending on James to bring the wine.

It is argued, morever, that despite the theoretical convergence cited that would preclude any raising treatment for (4), that is actually the correct analysis. If so, the theoretically ‘forbidden’ phenomenon exists in English. While many might be disturbed by controversy and lack of agreement in syntax, this essay could be taken to justify a certain skepticism about theoretical concord. Chapter 3, which originated as a lecture given at the University of Maryland in May 2000, deals with a curious set of restrictions initially visible in certain interactions with Complex DP Shift and Right Node Raising constructions and illustrated by contrasts like (5) and (6): (5)

a. I believe [the woman who is favored to win]1 to have screamed. b. I believe [the woman who is favored to win]1 to be a doctor. c. I believe [the woman who is favored to win]1 to be you.

(6)

a. I believe t1 to have screamed [the woman who is favored to win]1. b. I believe t1 to be a doctor [the woman who is favored to win]1. c. *I believe t1 to be you [the woman who is favored to win]1.

8

INTRODUCTION

Here while in general the arguably raised (to object) subject of an infinitival complement with a main verb like believe can be the target DP of the Complex DP Shift phenomenon, not so in cases like (5c). This chapter explores the reasons for this apparent anomaly and its connections to a number of other peculiar restrictions. There emerge grounds for skepticism both about views that see Complex DP Shift and Right Node Raising as fundamentally distinct from left extractions and about those that reject the notion of invisible resumptive pronouns, and there also emerges a straightforward argument for a ‘nonderivational’ property of these constructions. Chapter 4 argues for the existence of a grammatical category distinction in English DPs that both traditional grammar and more modern research seem mostly not to have recognized. This distinction is taken to underlie such a priori puzzling contrasts as (7): (7)

a. Tanya attended (in the matriculation sense) some school/*something/it. b. Tanya discussed some school/something/it. c. Tanya discussed some school 1/*something 1 after attending (in the matriculation sense) it1.

This chapter could be said to be skeptical of the possibility of an adequate account of English syntax that does not appeal to the posited category distinction. Chapter 5 grew out of research begun jointly with Haj Ross and took more form as a talk “On the Grammar of Squat” given at the Yale University Department of Linguistics Reunion on November 10, 2000. The skeptical morals here are severalfold. There is the general implication that much less is understood about the grammar of even well-studied NLs than is often implied. In this case, now quite current slang forms are shown to have hitherto largely unstudied and surprising properties relevant not only to their own analysis but also to that of central standard forms. Further, evidence emerges that even systematic morphological relations characteristic of a class of forms can be quite deceptive as to their implications for the grammatical analysis of those forms. The reader will notice that inter alia the topics of chapters 1–5, if developed as promised, will flesh out and support the claims at the beginning of this introduction that it is not serious to imagine that linguists are today in possession of, or even close to, adequate descriptive accounts of syntax, even for a heavily studied NL like English. Chapter 6, while raising serious issues of linguistic theory, is also in a sense transitional to the concerns of part II, as it uncovers an unpleasant feature of much common linguistic theorizing, namely, deep question begging. It is argued that a critical claim about the role of the lexicon that underlies the assumption that NLs could have generative grammars in the technical sense is supported in a literature that dates to the mid-1950s only by begged questions. This situation is not what one would expect to find in real linguistics but is entirely expected internal to the topic of part II, which is something specious that masquerades as linguistics. This chapter presents arguments based on a variety of phenomena that NLs have a property called openness, which precludes the possibility of their having proof-theoretic/generative grammars. The phenomena at issue include direct discourse.

INTRODUCTION

9

Part II contains eight essays, seven of which purport to uncover something that I claim is justifiably called junk linguistics. The latter is defined not by the substantive matters it touches on but by the standards, more accurately lack of same, that govern what to a significant extent only purports to be linguistic inquiry. It is one thing to lack insight, to propose defective principles, to suggest generalizations that do not stand up. All this is a regrettable but nearly inevitable part of normal inquiry. It is a quite different thing to flout minimal standards of scholarly procedure; to ignore the literature; to claim such and such generalization holds when one knows or should know it does not; to generalize to grand claims from a few selected cases; to develop techniques, rhetorical and otherwise, for avoiding falsification; to deliberately cite certain facts that support one’s proposals while deliberately not citing those that do not; to fail to respond to criticisms and to restate criticized positions as if no critique existed not because the challenges do not merit a response but because one lacks a viable response; to utilize other people’s ideas without credit; to claim that someone whose work one is criticizing has said such and such when there is no basis for such a claim4; and so on. Combinations of various of these and other unacceptable procedures inevitably yield something that, while purporting to be linguistics, is actually junk linguistics.5 The latter claims to advance the understanding of NL, but it is a pretense and fundamentally a deception.6 That junk linguistics should not exist is uncontroversial, and it would be soothing and thereby quite tempting to think that it does not. Moreover, the contemporary culture of linguists is, I believe, unwisely unreceptive to any claims that it does. Especially, this culture is certainly extremely defensive about questioning the standards of the influential and powerful. But facts about linguistics aside, this would be a puzzling state of affairs. For scientific fraud and fakery in general is hardly rare and whole works are devoted to its dissection; see for instance the excellent volume by Broad and Wade (1982).7 After all, the term junk science on which junk linguistics is calqued has a well-known nonnull extension. By what miracle could one expect that the field of linguistics would be free of such things? To give one concrete example, Postal and Pullum (1997) document a clear case where a major linguistic figure distorted to a biographer (the latter a person in general untutored in linguistics and its history) the history of that figure’s own department, a history in which I was marginally personally involved. The distortion, not surprisingly, made the figure appear far more positively with respect to the (remarkably marginal and minor) issues at stake than reality indicates. The validity of the historical claims in Postal and Pullum (1997) has never been challenged. But after it appeared, I received a note from the chairman of a major Department of Linguistics in the United States. Therein the chairman expressed distaste and fatigue with the sort of quarrelsome content of our article and expressed worries about its negative effect on students. No negative feeling was directed toward the dishonesty we had documented. That is, the chairman’s critical attitude challenged exclusively the revelation and correction of falsehood, not the falsehood itself, although nothing that addressed the conclusion of falsehood was presented. In my experience, this attitude, which sees exposure of misbehavior (at least that by a favored few) as in bad taste while ignoring or finding some strained justifcation for the behavior itself is not isolated. But a field that allows the pseudovalue of consensus that includes at

10

INTRODUCTION

least a dictated silence about misdeeds primacy over the truth, some of which may be displeasing, is all too likely to obtain a lot more consensus than truth. Junk linguistics of course attempts to impersonate real linguistics as junk science in general attempts to impersonate real science. My view is that the best defense against junk linguistics is precisely the skeptical attitude referenced in the title of this work. Here there are several aspects. First, one must of course intellectually challenge, at least to oneself, claims that such and such generalization or princple holds either for a particular NL or, more grandly, for all. One wants to know where this has been shown, what data sets support it, what data sets if any attack it, and so on. When one is told that so and so has shown such and such, one needs to demand references, specifically page references.8 Second, one must try to analyze as clearly as possible the logic of the purported support for various proposals. However, certain aspects of junk linguistics are more insidious than simply inadequately supported or false claims. A recurring feature is utilization of a pretentious terminology to make grandiose-sounding claims that, on analysis, turn out to involve no factual propositions about the subject matter at all. Good examples of this with respect to the study of passivization are found in chapter 8. One different example should suffice for illustration here. Smith (1999: 1) begins his investigation of the linguistic contributions of a leading contemporary figure by stating flatly that “he has shown that there is really only one human language.” This claim is then repeated (1999: 214) at the end of Smith’s laudatory volume, indicating that it is meant to represent an important accomplishment. Notable is that nowhere is there any citation of a work or even part of a work where a demonstration of this seemingly remarkable property could be found. Nor is there any even attempted analysis of what the claim of ‘only one language’ could mean. Evidently, it is not intended to signify that we all speak Mohawk.9 What could show that it is false and that there are, say, two, nine, or sixteen hundred languages? Given the undeniably vast diversity among the totality of attested NLs, such a claim can only mean that there are certain similarities among them. The situation is entirely analogous to that for the totality of known human beings, which reveals a vast diversity and yet an enormous core of common features (hence, e.g., the possibility of medical science). But is any biologist going to claim thereby that there is only one human being? Evidently then, the relevant claim is empty, mere use of an unanalyzed terminology (‘one language’) with presumed deep and important connotations to cover a claim no less devoid of content than would be a ‘one human being’ claim. Such empty puffery is typical of junk linguistics. Especially disturbing is that such vacuous ‘accomplishments’ can be claimed, discussed, published, reviewed, and so forth, all without any outcry, as if the difference between such pretense and real inquiry was not noticed.10 This should be a warning that the rot of junk linguistics is widespread and little recognized. But another aspect of the invocation of contentless accomplishments like the ‘one language’ claim deserve stress. Research that has real scientific accomplishments has no need to cite such empty pseudoresults. Their invocation should be regarded, first, as an attempt to fill the gap left by the lack of actual results and, second, consequently, in effect as an unintended admission of the lack of same. Remarkable then is the apparently widespread inability to hear the admission.11

INTRODUCTION

11

Relevant to an understanding of this situation, no doubt, are the mind-clouding effects of (i) the repetition of claims, (ii) the association of claims with prestigeful figures, and (iii) the creation by the process of education of interest groups whose career success depends as much or more on the conformity of expressed views with certain doctrines as on their validty and scientific grounding. Here one should, for instance, look most carefully at the recent discussion of the introduction and widespread acceptance of the so-called minimalist program, arguably successfully introduced with no groundings in descriptive success whatever.12 The fact that one finds a claim repeated in ten, twenty, or one hundred works cannot, no one doubts, in itself legitimize the claim. But there is nonetheless a tendency to attribute seriousness to widely cited claims. My experience suggests, though, that wide affirmation of a claim is in current linguistics not even ground for confidence that it is serious, still less that it is well supported or true. As already indicated, chapter 2, though not concerned with junk linguistics, documents a case of broad crossframework agreement that a certain syntactic phenomenon, arguably found in English, is universally impossible. Chapter 7 examines a popular and widely cited purported explanation of the strong crossover effect, a phenomenon apparently first made explicit in Postal (1971). The junk linguistic character of this proposal is revealed sharply inter alia by the fact that the so-called explanatory account can be shown to fail not on the basis of data from some exotic NL but merely because of English data, already published (in Postal, 1971) a decade before the essential junk claim appeared. Ominously, this state of affairs in no way interfered with a long and happy acceptance of the junk claim. That is, for a junk linguistic proposal, having been, in effect, already shown to be false before it was even formulated appears to be no current barrier to success. Chapter 8 considers a range of transformational grammar views about English passives. It is argued that they manifest a characteristic combination of junk features that includes seemingly strong claims actually rendered empty by rhetorical dodges, grotesque factual incompatibility of claims if interpreted in a way with factual content, and an overall deep contempt for the facts, standards of argument, the literature, and readers. It is shown, for instance, that some strong (if interpreted factually) transformational claims about passives of the 1970s were not only already known in the 1950s to be false but also known to be such by the author of the later claims. Chapter 9, previously published in 1988, takes the form of a series of parodies of junk linguistic claims, although the term ‘junk linguistics’ was not used. But the parody form should not obscure the serious intent. Herein is cited inter alia what is, one hopes, the only case in the history of inquiry where an author states that a proposition A is a theorem of certain assumptions B, then states that he has no proof of A from B, and then adds that it is unlikely any proof can be constructed. A twentyfirst-century reader might imagine that such incoherence could, at worst, only have appeared outside of a parody in some marginal outlet, a very bad beginning student paper, a set of parochial working papers, or a missupervised thesis in, maybe, a deprived or isolated third world university, and so on. Not at all; it was published in what many might consider the then (and now) most prestigious current journal of theoretical linguistics. Moreover, despite the absurdity, in the intervening twentyfour years, neither the author nor the editor who approved its publication have, to

12

INTRODUCTION

my knowledge, in any way recognized the junk linguistic character of the passage, which thus remains unrepaired.13 Except for its place in my parody and a citation in Postal (1982), I do not know that it is ever mentioned. The implication is, evidently, that even the most obvious junk linguistic failure, one trivially discernible independent of facts and entirely on internal grounds, is cost-free for its perpetrator. Anyone who doubts that junk linguistics is a real and longstanding phenomenon should try to find an alternative account for all of (i) the production and writing of the set of claims just cited, (ii) their unimpeded passage before the eyes of several readers for this prestigious journal, (iii) their satisfaction of the editor’s standards for publication, and (iv) their survival over decades without correction. A key conclusion is that, unlike the parallel situation with junk work in many other fields, junk linguistics seems to incur no sanctions. Chapter 10 argues for the intrusion of junk linguistics into the refereeing process for the National Science Foundation. A specific referee report is analyzed in detail and argued to be a nearly perfect model of junk linguistic thinking disguised as refereeing. Chapter 11 focuses on issues of ethics and honesty in a passage putatively concerned with the foundations of linguistics. Here is junk linguistics arguably at its absolute worst with an author contradicting himself internal to the passage, contradicting other nearly contemporary work of his own, contradicting his own accurate remarks of decades earlier, scorning fundamental and universally acclaimed results of modern logic, and ignoring criticisms of his position he evidently cannot answer, all in aid of a foundational position that is straightforwardly incoherent. The reader who for whatever reasons a priori doubts the existence of junk linguistics might well be advised to begin here. Chapter 12 undertakes a bit of lexicographic analysis of arguably curious and obscure usages of linguists themselves, ones like “X follows automatically from Y,” “X virtually holds,” and “X is natural.” It is suggested that these usages in themselves reveal a strong underlying current of junk linguistic thinking. Chapter 13, the final substantive essay of part II, investigates recent claims that such and such supposed feature of NL is (virtually) conceptually necessary. It is suggested that such jargon is a substantively empty means, hence one to be expected in junk linguistics, of promulgating claims in the absence of any serious argument or evidence for them. It is argued in detail that certain claims of this type are not only unsupported but also unsupportable, since they are false. A key issue involved is what amounts to rhetorical facilitation of serious and very longstanding question begging about the nature of NL grammars. Chapter 14 provides a brief set of schematic personal suggestions of how one should in principle deal with junk linguistics. As signaled in this book’s title, a key component should be a thoroughgoing skepticism, one that begins with one’s own views. Those who have different ideas about how to react to junk linguistics should be urged to make them public.

I

STUDIES IN LINGUISTICS


1

A Paradox in English Syntax

1. Introduction 1.1. Paradoxes The notion of ‘paradox’ relevant to this essay is nicely defined in (1): (1)

Sainsbury (1988: 1) “This is what I understand by a paradox: an apparently unacceptable conclusion derived by apparently acceptable reasoning from apparently acceptable premises. Appearances have to deceive, since the acceptable cannot lead by acceptable steps to the unacceptable. So, generally, we have a choice: Either the conclusion is not really unacceptable, or else the starting point, or the reasoning, has some nonobvious flaw.”

Early sections of this study argue that common, perhaps even standard assumptions about English grammar yield a paradox in this sense when combined with certain observations about a puzzling class of sentences. It is ultimately claimed that one can resolve the paradox by adopting a particular set of grammatical assumptions; a partial development of these makes up section 9. 1.2. Factual background A characteristic feature of one rather formal style of English is the existence of structures I refer to with considerable theoretical prejudice as null expletive subject (NEX) clauses. They are illustrated by the sentences in (2): 15

16 (2)


a. Under the table was lying an elderly crocodile. b. Toward the island advanced the huge enemy armada. c. Aissen and Hankamer (1972: 502) Under this verb is embedded the S containing the cyclic subject. d. Aissen and Hankamer (1972: 502) To these causes are attributed most of the financial catastrophes of the decade. e. Aissen and Hankamer (1972: 502) To the social director fell the task of finding accommodations for all the visiting icthyologists. f. Green (1985: 117) To every VP rule with certain properties corresponds an S rule with related properties.

A rough atheoretical characterization of NEX clauses is as follows: They consist of a preverbal PP, henceforth the X-PP, a verbal expression, possibly other elements, and a postverbal DP expression that determines the finite verb agreement. Hereafter, I refer to this DP as the P(otential)-subject, since, in general, each NEX clause corresponds to a non-NEX clause where the P-subject is in standard subject position (and the X-PP possibly postverbal). NEX clauses were probably first analyzed in modern terms in Emonds (1970, 1976). NEX clauses have rather curious properties linked to the notion subject just mentioned as follows. Most English (finite) clauses have an unambiguously identifiable element one can refer to as a standard superficial subject (hereafter S3). The S3, typically a DP,1 precedes the finite verb in nonsubject/auxiliary inversion clauses, determines verb agreement, concords with a tag if one is present, corresponds to the raised element in a raising construction with verbs like seem, occurs postposed to the first auxiliary in subject/auxiliary inversion clauses, and so on. It is plausible to suggest, at least for English, that principle (3) holds: (3)

Every finite clause has one and only one S3.2

NEX main clauses challenge (3) in that there is no clear candidate for S3 status. To this challenge there are in effect several logical responses and various proposals actually found in the literature. One response would simply reject (3) and take NEXs to have no S3s. If though (3) is maintained there are at least three subpositions with respect to NEXs, represented in (4): (4)

a. The S3 is an invisible expletive; see Coopmans (1989), Lawler (1977), and Postal (1977). b. The S3 is the postverbal DP that detemines verb agreement, that is, the P-subject (e.g., the huge enemy armada in (2); see Bresnan (1976, 1977), Kathol and Levine (1993), and Levine (1989)). c. The S3 is the X-PP.

Although (4c) might, given widely accepted assumptions about the nature of S3s, a priori seem far-fetched, such a position was proposed in Stowell (1981: 269–276) and accepted in Pesetsky (1981/1982: 330). Moreover, in a number of works, Bresnan

A PARADOX IN ENGLISH SYNTAX

17

(1989, 1991, 1994, 1995) and Bresnan and Kanerva (1989, 1992) consider all of (4), present arguments against (4a), and conclude that (4c) is the correct analysis of NEX clauses. More precisely, this is concluded about the restricted subclass of NEX clauses dealt with, those where the fronted phrase is a locative or directional PP, which Bresnan references under the rubric locative inversion. As mentioned later, this terminology is hardly ideal, as there is a wide range of otherwise structurally similar NEX clauses whose elements are not subsumed by even the broadest interpretation of ‘locative’. Bresnan’s conclusions were based inter alia on two important observations she has made about the interactions of NEXs with extraction constraints. These appear to show that the X-PPs in NEXs obey constraints on extracted subjects, that is, on extracted S3s. This evidence is strengthened later. But the burden of this essay is to argue that rather than justifying conclusion (4c), Bresnan’s work in effect reveals a paradox in English syntax. This arises when one considers (i) certain common assumptions about English grammar, (ii) the strengthened evidence from extraction facts that seemingly indicate that the X-PPs in NEX clauses are subjects, and (iii) a variety of evidence, which Bresnan did not consider, that X-PPs are not subjects. As with other intellectual paradoxes, the recognition of this one is ultimately not a negative result but rather a genuine opportunity to understand something deeper about the subject matter. For if, as it turns out, given a background set of assumptions A, different factual assumptions suggest both that certain phrases in certain sentences have and do not have certain properties P (here, those defining subjecthood in a specific sense), at least one of A must be false. So I agree with Bresnan (1994) that NEX clauses tell us something important about the ‘architecture of universal grammar’, as well as about English grammar in particular, but for very different reasons. There is little overlap between the conclusions I draw and those induced in Bresnan’s recent work. Where the latter concludes that NEX cases support the version of the Lexical Functional Grammar (LFG) framework she advocates, I will appeal to a quite different set of assumptions. As already mentioned, the class of NEX clauses on which Bresnan focuses are those commonly and not unnaturally called locative inversion cases. ‘Locative’ in this sense is characterized by Bresnan (1994: 75) as subsuming “a broad range of spatial locations, paths, or directions, and their extensions to some temporal and abstract locative domains.” This seems to me an inadequate and rather arbitrary way to slice up the class of NEX clauses. Even taking into account the vague and unclarified ‘extensions’ and ‘abstract’ locative domains, this kind of description does not give a real sense of the richness of the class of PPs that define NEX structures.3 That characterization seems, for instance, to have no obvious application to many relatively banal NEX examples: (5)

a. Lawler (1977: 502) To the states is entrusted the power to regulate education (by the Constitution). b. For that perverted cause were slaughtered thousands of innocents. c. From condition (i) can be deduced the fact that every verb contains two vowels. d. During the reign of Queen Lulu II were built many fabulous monuments. e. With this pen seems to have been written the first verse of that famous sonnet. f. Against that proposal can be objected the fact that no one is entirely logical.

18


g. Throughout that period were undertaken some impressive feats of irrigation. h. To those questions correspond the following answers. i. On this election may well depend the future of our entire planet.

Here, at least examples (5a, b, c, e, f, and h) do not seem to be subsumed by Bresnan’s characterization. Nonetheless, the basic properties that define what appears to me to just be the locative inversion subpart of the NEX construction are equally found in structures that do not involve locatives, like those of (5); these properties include, as will be shown, the extraction facts that loom so large in Bresnan’s conclusion that the X-PPs of NEX clauses are (functional) subjects. It is not clear that there is any English-internal justification for picking out locative inversion cases as a separate construction; in any event, I do not do so here.

2. Subject properties of X-PPs 2.1. Remarks Although Bresnan (1976, 1977) assumed (4b),4 her recent works adopt assumption (4c), as specifically interpreted internal to the LFG framework. Under this interpretation, the X-PPs in NEX clauses are subjects at the (grammatical) function level (functional structure) but not phrase structure subjects at the surface level, since in effect they have been extracted. I certainly agree with Bresnan as well as other work that the X-PPs are extracted. In cases like (2), the extraction can presumably be identified with PP topicalization. But although the bulk of work on NEX clauses seems limited to PP topicalization cases, otherwise parallel clauses manifest other types of extraction, highlighted in (6):5 (6)

a. Aissen and Hankamer (1972: 502) They are planning to destroy the old church, under which are buried six martyrs. b. Aissen and Hankamer (1972: 503) These are the causes to which are attributed most of the financial catastrophes of the decade. c. On what wall should be hung the portrait of Generalissimo Francisco Franco? d. During what period were first developed effective pork-preservation methods? e. It was there that were first developed effective pork-preservation methods. f. No matter where he thinks were first developed pork-preservation methods, . . .

That X-PPs are extracted, a point on which Bresnan and I are in total agreement, thoroughly undermines any argument for the subjecthood of X-PPs based on subject raising, that is, based on facts like (7): (7)

a. b. c. d. e. f.

On the wall were standing two large blackbirds. On the wall seemed to be standing two large blackbirds. On the wall I believe to have been standing two large blackbirds. On the wall are believed to have been standing two large blackbirds. *I believed on the wall to have been standing two large blackbirds. *It is impossible for on the wall to have been standing two large blackbirds.


19

For, as partially noted in Lawler (1977: 234–235), this array of facts is what would be expected under view (4); in particular, in each grammatical case, the X-PP can be analyzed as being in an ordinary extraction position. Notably, (6e and f), where this is not the case, are ungrammatical, which can be attributed to the well-known incompatibility of infinitival clauses with domain internal extraction. Given these longknown facts, the claim of Bresnan (1994: 96) that “inverted locative phrases undergo subject raising (Postal 1977)” is unfounded, since such an assertion cannot be supported merely on the basis of data like that in (6) and Bresnan provides no other grounds. While the facts in (6) are consistent with the view that their X-PPs are subjects, they are no less consistent with the view that an invisible expletive is the raised subject. Hence such facts play no known role in differentiating the viability of views (4a) and (4c). Arguments like that of Pesetsky (1981/1982: 330) simply do not go through: “As Joan Bresnan and Donca Steriade have pointed out to me, the idea that these PPs and adverbials move to subject position is supported by the fact, noted by Postal (1977) that these elements can undergo Subject-to-Subject raising.”6 If, however, raising facts like (6) contributes nothing to arguing for the subjecthood of X-PPs in NEX cases, Bresnan has developed two rather striking and unexpected types of argument that seemingly do. 2.2. The Anticomplementizer Constraint argument One goes back to the observation in Bresnan (1977: 186) that, as Bresnan (1994: 97) states it: “[T]he preposed locatives in locative inversions show the constraints on subject extraction adjacent to complementizers”: (8)

a. It is in these villages that we all believe __ can be found the best examples of this cuisine. b. *It is in these villages that we all believe that __ can be found the best examples of this cuisine.

Thus the X-PPs in NEX structures seem to manifest the so-called that-trace effect, otherwise linked only to subject extraction. So *(8b) behaves like the regular subject extraction case *(9a) and not like the regular adjunct extraction case (9b): (9)

a. *It is these villages that we all believe that __ contain the best examples of this cuisine. b. It is in these villages that we all believe (that) the best examples of this cuisine can be found.

I refer to facts like those in (9a) as the Anticomplementizer Constraint. Bresnan’s widely neglected discovery that the X-PPs in locative inversion NEX clauses manifest the Anticomplementizer Constraint is a striking argument for the subjecthood of those PPs, given the not heretofore seriously challenged (but see later text) assumption that in all other English cases the extracted elements that manifest the Anticomplementizer Constraint are S3s. Before continuing, it is worth observing that the Anticomplementizer Constraint holds as well for NEX clauses that do not fall under the locative inversion rubric:

20 (10)


a. To these causes I believe (*that) are attributed most of the financial catastrophes of the decade. b. To the social director I believe (*that) fell the task of finding accommodations for all the visiting icthyologists. c. During the reign of Queen Lulu II I believe (*that) were built many fabulous monuments. d. On this election I believe (*that) may well depend the future of our entire planet.

2.3. The Parallelism Constraint argument The Anticomplementizer Constraint argument alone is a serious basis for Bresnan’s adoption of conclusion (4c). And it takes on greater force in combination with another rather parallel argument she has developed. This involves what she calls the Parallelism Constraint on across-the-board extraction from coordinate constituents. This constraint (see Falk, 1983; Gazdar, 1981; and Woolford, 1987) is illustrated by Bresnan (1994: 98) with data like (11): (11)

a. b. c. d.

She’s someone that __loves cooking and __ hates jogging. She’s someone that cooking amuses __ and jogging bores __. *She’s someone that cooking amuses __ and __ hates jogging. She’s someone that cooking amuses __ and I expect __ will hate jogging.

According to Bresnan, the generalization is that “subject gaps at the top level of one coördinate constituent cannot occur with any other kind of gap in the other coördinate constituent.” This generalization distinguishes (11a), in which both the gaps correspond to subjects, and (11b), in which they both correspond to objects, from the bad (11c), in which a subject gap is mixed with a nonsubject gap. (11d) is then nonetheless well formed, because the subject gap is not at the highest level of the conjunct that contains it.7 Bresnan then goes on to note insightfully that the X-PPs in locative inversion cases obey the Parallelism Constraint, citing such data as (12): (12)

Bresnan (1994: 100) a. That’s the old graveyard, in which __ is buried a pirate and __ is likely to be buried a treasure. b. That’s the old graveyard in which workers are digging __ and a treasure is likely to be buried __. c. *That’s the old graveyard, in which workers are digging __ and __ is likely to be buried a treasure. d. That’s the old graveyard, in which workers are digging __ and they say __ is buried a treasure.

As Bresnan observes, the pattern in (12) matches that in (10), which would follow from the Parallelism Constraint if the X-PPs are subjects, as (4c) claims.8 As with the Anticomplementizer Constraint argument, the properties Bresnan has documented are not limited to the locative inversion subset of NEX clauses:


(13)

21

a. That is the period during which were built many large monuments __ and were proved to have been fought several fierce battles. b. That is the period during which many large monuments were built and several fierce battles were proved to have been fought. c. *That is the period during which many large monuments were built and were proved to have been fought several fierce battles. d. That is the period during which many large monuments were built and they say were proved to have been fought several fierce battles.

One might accept, though with great reluctance, that either the Anticomplementizer Constraint argument facts or the Parallelism Constraint facts for NEX cases are some kind of fortuitous accident. But that both constraints hold for NEX cases as well as standard subject cases renders recourse to such a position at the least highly implausible. 2.4. The Complementizer Effect argument Moreover, there is in effect a third argument from extraction facts parallel to Bresnan’s that also seems to support the claim that X-PPs are subjects. This is based on complementizer facts found inter alia in relative clauses, including those occurring in cleft structures. The generalization is that when a subject is extracted in the absence of a relative pronoun, the complementizer that is obligatory, but not so when nonsubjects are extracted. Since this is something of the opposite of the situation described by the Anticomplementizer Constraint, call it the Complementizer Effect: (14)

a. They were discussing the spaceship that/Ø the scientists built. b. They were discussing the spaceship that/*Ø attacked the town.

Clefts are then crucial because they permit the extraction of PPs. Notably then, the Complementizer Effect is found when the a clefted constituent is the X-PP of a NEX structure, although not when it is an arbitrary PP: (15)

a. b. c. d. e.

It was those towns that/Ø he studied. It was those towns that/*Ø were studied. It was those towns that/Ø she talked about. It was in those towns that/Ø she learned the best techniques for drying fruit. It was in those towns that/*Ø were learned the best techniques for drying fruit.

2.5. The Adverb Effect The correctness of Bresnan’s insight that the extracted PPs of NEX clauses obey the Anticomplementizer Constraint is corroborated by what work of Culicover (1991, 1993a, 1993b) refers to as the Adverb Effect. Roughly, Culicover notes that in the presence of certain fronted adverbials the Anticomplementizer Constraint is not found. Significantly for our purposes here, Culicover (1993a: 104) notes that the same nullification of the Anticomplementizer Constraint by a fronted adverbial occurs in NEX cases:9

22 (16)


a. Standard Anticomplementizer Constraint *Robin met the man who1 Leslie said that t1 was the mayor of the city. b. Adverb Effect (Culicover, 1993a: 98) Robin met the man who1 Leslie said that for all intents and purposes t1 was the mayor of the city. c. Standard Anticomplementizer Constraint in NEX Case (Culicover, 1993a: 104) *[On which table]1 were you wondering whether t1 had been put the books that you had bought d. Adverb Effect in NEX Case (Culicover, 1993a: 104) [On which table]1 were you wondering whether under certain circumstances t1 might have been put the books that you had bought

These parallelisms further strengthen Bresnan’s original observation and indicate beyond much doubt that X-PP extraction in NEX cases is governed by the Anticomplementizer Constraint. 2.6. Summary It appears then that the X-PPs of NEX clauses obey the three extraction constraints, the Anticomplementizer Constraint, the Parallelism Constraint, and the Complementizer Effect. This is a substantial basis for concluding that (4c) is correct and provides serious motivation for Bresnan’s recent adoption of that position.

3. Nonsubject properties of X-PPs 3.1. Remarks However, while a body of real evidence that argues for the subject status of the X-PPs of NEX clauses has been assembled, this is by no means the end of the matter. For except for consideration of the presence of a P and the failure to determine finite verb agreement, two very nonsubjectlike properties of these putative PP subjects, Bresnan’s work that attempts to justify (4c) has been largely inattentive to a number of other discernible ways in which the putative PP subjects fail to behave like subjects (S3s). This is true, I believe, despite her remark (1994: 97, note 36) that indicates the change of position represented by Bresnan (1994): “Since inverted locatives show the same effect as subjects, but differ in some other respects from subjects, Bresnan 1977 concludes that the extraction constraint cannot reflect syntactic subject status.” When, however, a collection of such properties is attested, it becomes apparent that an immediate conclusion from the materials of section 2 that X-PPs are subjects in the same sense as standard S3s is premature at best. Let us then consider nonsubject features of the putative subjects. 3.2. Failure to determine finite verb agreement Failure to determine finite verb agreement renders the forms anomalous under a subject analysis, since there seem to be no other attested S3s with this property.


23

3.3. PP subjects The presence of a P or, equivalently, the fact that the putative subjects are PPs is also an anomaly, since again no other attested S3s seem to have this property, either. I agree with Bresnan (1994: 110–111) that examples like (17) that involve initial PPs and predicate nominals are consistent with this claim: (17)

a. b. c. d.

Under the table is a good place to put the box. Under the table seems to be a good place to put the box. He believes under the table to be a good place to put the box. Is under the table a good place to put the box?

For, as (17d) shows, these apparent PP subjects, unlike the X-PPs of NEX cases, can appear in postauxiliary position and also in clear nonextracted positions like that of (17c). I agree with Bresnan’s suggestion that cases like (17) most likely involve some form of invisible DP head and noun, so that (17a) would have a subject DP of schematically the form [a place under the table].10 Beyond the support for this view given in Bresnan (1994: 110–111), consider: (18)

a. [Under the table]1 looks/seems like it1 is a good place to put the box. b. *[Under the table]1 looks/seems like it1/there1/Ø1 was placed the missing box.

That is, the phrase in the construction of (17) can, like an uncontroversial S3, antecede a resumptive pronoun in the look/seem + like construction, returned to briefly in chapter 3, but the X-PP of a NEX case cannot. 3.4. Standard floating quantifiers The proper type of S3 can, of course, link to standard floating quantifiers (SFQs), all, both, and each, found in VP-internal positions, an ability that is unaffected when the subject is extracted. But X-PPs can never link to SFQs:11 (19)

a. b. c. d. e.

Those women have all/both/each filed a complaint. Those women, I am sure __ have all/both/each filed a complaint. To those women were (*all/*both/*each) proposed a distinct alternative. Under those chairs I believe have (*both) been found that kind of dust. During their reigns were (*each) annexed a wealth-producing adjacent territory.

In further support of the contrast between X-PPs and PP subjects like those in (17), the latter can link to SFQs: (20)

a. Under the table and under the bed would both be good places to store our ski equipment. b. Under the table and under the bed was/were (*both) stored our ski equipment.

This further attacks the subjecthood of X-PPs.

24


3.5. Nonstandard floating quantifiers By nonstandard floating quantifiers (NFQs) I refer to those instances of each that are semantically linked to one NP but occur postposed (possibly cliticized) to another, as in (21): (21)

He sent those girls three photos each.

NFQs can, of course, link to S3s, even when these are extracted: (22)

a. Those girls, I am sure __ sent him three photos each. b. It was those girls who I am sure __ sent him three photos each. c. It was those girls who I am sure __ were sent three photos each.

As (22c) indicates, S3s that would be regarded as ‘derived’ in transformational terms can also link to NFQs. Notably, though, NFQs never link to X-PPs:12 (23)

a. Under those tables were sitting two frogs (*each). b. To the two officers fell three complicated tasks (*each). c. At those tanks were fired three rockets (*each).

Compare the following contrast between an uncontroversial S3 and an X-PP based on lexically and semantically identical DPs with the same verb: (24)

a. Those candidates (he learned) had been sent three questionnaires (each). b. To those candidates (he learned) had been sent three questionnaires (*each).

3.6. Floating emphatic reflexives Standard S3s can link to floating emphatic reflexives (FERs), a possibility that is also unaffected when the subject is extracted: (25)

a. Those women have themselves filed complaints. b. Those women, he believes __ will soon themselves file complaints.

But X-PPs never link to FERs, contrasting again with the cases of (17): (26)

a. To those women were (*themselves) given the opportunity to resign. b. Under that sofa may (*itself) have been lying two snakes. c. Under the table may (itself) have been a good place to hide a snake.

3.7. Controllers S3s can, of course, function as the antecedents for control, that is, can be par excellence controllers for subjects of nonfinite adjuncts, for example:


(27)

25

a. [Those women]1 were given the opportunity to file complaints without Ø1 having to reveal their identities. b. [The two men]1 stood near Jane and Clarissa after Ø1 entering the room.

But this is in general not possible for X-PPs: (28)

a. *To [those women]1 was given the opportunity to file complaints without Ø1 having to reveal their identities. b. *Near [Jane and Clarissa]1 stood the two men after Ø1 entering the room. c. *At the fortress1 were fired seven rockets while Ø1 being attacked by tanks. d. [The chimp]1 was handed a banana without Ø1 being handed a peach. e. *[To the chimp]1 was handed a banana without Ø1 being handed a peach.

In this respect, X-PPs again contrast with the sort of true PP S3s of cases like (17), since, as noted in Hornstein (2001: 78), the latter can be controllers: (29)

[In the bathroom]1 is a great place to hide without Ø1 really being a good place to live.

3.8. Reflexive/Reciprocal X-PPs English S3s cannot be reflexives or reciprocals bound by other elements of their clause. Most theories of reflexives are designed to yield this as a consequence; see, for example, Chomsky (1981) and Pollard and Sag (1992, 1994). Call this the Reflexive/Reciprocal S3 Constraint. This constraint is unaffected by extraction of the offending form:13 (30)

a. *Himself1 discussed Mike1 with the other students. b. *It was himself1 that __ discussed Mike1 with the other students. c. *Himself1, they said __ discussed Mike1 with the other students.

Consider then: (31)

a. *To Mike1, himself1 should never be described. b. ?To himself1, Mike1 should never be described. c. ??To himself1 is said to have been unexpectedly described the only guy1 who thought he was handsome.

The point here is no doubt subtle. While (31c) is hardly acceptable, its quality seems much closer to the less than perfect non-S3 case (31b) than to the totally impossible Reflexive/Reciprocal S3 Constraint violation (31a). A less fuzzy paradigm is seen in (32): (32)

a. Those tapes seem to have been bought by Ed1 for himself1. b. ?For himself1 three of the tapes in question seem to have been bought by Ed1. c. For himself1 seem to have been bought by Ed1 three of the tapes in question.

26


For me, notably, (32c), which would be a Reflexive/Reciprocal S3 Constraint violation under the proposal that X-PPs are subjects, is, if anything, better than (32b). The reciprocal facts are also relatively clear: (33)

a. b. c. d.

Those guys1 bought such things for each other1. For each other1 those guys1 would never buy such things. ?For each other1 such things would never be bought by those guys1. ?For each other1 would never have been bought by those guys1 the things in question. e. *Each other1 would never buy such things for those guys1.

Here (33d) does not seem appreciably worse than (33c) and so it lacks the utter impossibility of a Reflexive/Reciprocal S3 Constraint violation such as *(30a) or *(31a) or *(33e). See also: (34)

a. b. c. d. e. f.

Those girls1 would never show such things to each other1. *Each other1 would never show such things to those girls1. *Each other1 would never be shown such things by those girls1. ?Those girls1 would never be shown such things by each other1. ?To each other1 such things would never be shown by those girls1. ?To each other1 would never be shown by those girls1 the sort of things in question.

The bottom line is that none of (31c), (33c), (33d), or (34f) is anything like as bad as a standard Reflexive/Reciprocal S3 Constraint violation. 3.9. Reflexives/Reciprocals anteceded by X-PPs Consider now something of the reverse of the situation in section 3.8, namely, where reflexives or reciprocals are anteceded by X-PPs of NEX clauses. Since hypothesis (4c) claims X-PPs are S3s and S3s are the best possible reflexive or reciprocal antecedents, at first glance the ungrammaticalities in (35) seem to count against the position: (35)

a. Sally and Louise were described to themselves/each other. b. To Sally and Louise were described the two doctors/*themselves/*each other.

However, this is not obvious since a defender of (4c) might attribute these to the independent fact that P-subjects in NEX clauses cannot be definite pronouns. So the bad cases of (35b) would be so for the same reasons, for example, as (36): (36)

*To Sally and Louise were described you/them/us.

But even if this move successfully defends (4c) against cases like (35b), it does not work in general. For parallel reflexive/reciprocal facts involve conjoined DPs and there exists no ban on a P-subject as a conjoined DP one of whose conjuncts is a pronoun; see Iwakura (1978: 329). Thus facts like (37) really do argue against the S3-hood of X-PPs:


(37)

a. b. c. d. e.

27

Sally and Louise were described to Mike and themselves/each other. Sally and Louise were described to themselves/each other and Mike. To Sally and Louise were described Mike and us/you/them. *To Sally and Louise were described Mike and themselves/each other. *To Sally and Louise were described themselves/each other and Mike.

Observe that the ill-formed (37e) is parallel to (38c): (38)

a. Mary talked about Sally/*himself to Fred. b. *Mary talked about Sally and himself to Fred. c. *To Fred, Mary talked about (Sally and) himself.

3.10. Unembedded subject topicalization Under position (4c) advocated in Bresnan (1994), an example like (39) involves a topicalized PP subject: (39)

Above the ranch was hovering a standard type of flying saucer.

However, while standard S3s are capable of topicalization, as in (40b), there is no reason to believe that in (40a) the first NP can, still less must, be analyzed as topicalized: (40)

a. A standard type of flying saucer was hovering above the ranch. b. That type of flying saucer, they never learned was hovering above the ranch.

That is, there seems to be a gap in topicalization paradigms that corresponds to an S3 of a main clause. Moreover, the intuitive view that cases like (40a) do not involve any kind of extraction is supported by a telling argument developed in Lasnik and Saito (1992: 110–111) on the basis of certain reflexivization facts. These authors argue specifically that while ‘long’ topicalization of a subject is possible, as in, for example, (40b), ‘short’ topicalization of a subject is impossible. Their strongest argument for this conclusion depends on paradigms like (41): (41)

a. b. c. d.

*John1 thinks that Mary likes himself1. John1 thinks that himself1, Mary likes. *John1 thinks that himself1 likes Mary. *John1 thinks that himself1, likes Mary.

The argument runs as follows: Topicalization ‘rescues’ certain otherwise impossible reflexives, as in (41b). But such examples involve object topicalization. The parallel subject cases are illustrated by (41c and d). As Lasnik and Saito observe, if ‘short’ topicalization of subjects were possible (necessary), then the principles that allow (41b) should also allow (41d). Since the latter are barred, the hypothesis that ‘short’ subject topicalization exists should be rejected. Given that, the fact that (4c) requires obligatory topicalization of its puta-

28


tive subject PPs shows that in still another respect the putative subjects of that analysis are anomalous. 3.11. ‘Extraposed’ phrases 3.11.1. ‘Extraposed’ relative clauses A property of uncontroversial English S3s is that they permit ‘extraposition’ of relative clauses, however one analyzes that theoretically. This phenomenon is entirely compatible with extraction of the relevant S3. But the X-PPs of NEX clauses, putative S3s under hypothesis (4c), never permit this: (42)

a. b. c. d.

[What woman]1 did they claim was t1 served beer (who1 had ordered wine)? [To what woman]1 did they think t1 was served beer (*who1 had ordered wine)? [What bridge]1 t1 collapsed (that had cost over a billion dollars to build)? [Under what bridge]1 did they think t1 lived a troll (quite happily) (*which had cost over a billion dollars to build)?

This also counts against the view that X-PPs are subjects in the sense of S3s. 3.11.2. ‘Extraposed’ exceptive phrases Parallel to relative clause ‘extraposition’, uncontroversial English S3s permit ‘extraposition’ of ‘exceptive’ phrases, regardless of whether the S3 is extracted. But again, this is impossible with the X-PPs of NEX clauses: (43)

a. [Who else except me]1 are you sure t1 was sent candy? b. [Who else]1 are you sure t1 was sent candy (except me)? c. [To who(m) else]1 are you sure t1 was sent candy (*except me)?

3.11.3. Summary I hope to have shown then that there is a range of evidence that clearly counts against the view that X-PPs are S3s.

4. The paradox: X-PPs as both subjects and nonsubjects Within the perspective developed by Bresnan (1989, 1991, 1994, 1995), the extraction arguments of section 2 show that the X-PPs of NEX cases are subjects. The limited nonsubject behavior of X-PPs that she deals with can then be treated as minor or peripheral features to be explained in various ways. Thus Bresnan (1994: 117–118) attributes the failure of X-PPs to manifest the normal S3 determination of finite verbal agreement to the absence of certain syntactic features on PPs. Whatever plausibility that general strategy had given the limited range of nonsubject properties of X-PPs mentioned by Bresnan, it has essentially none given the system-


29

atic failure of X-PPs to manifest the numerous subject (S3) properties cataloged in the previous section. It is not just a question of finding some alternative accounts for the lack of these S3 properties. Even (highly implausible) success in that venture would not obscure the fact that the supposed X-PP subjects of analysis (4c) simply do not behave like uncontroversial S3s in many clear ways. This collection of facts forms a paradox when combined with the extraction evidence assembled by Bresnan and strengthened in the previous section to the effect that X-PPs are subjects. Reversing Bresnan’s logic also leads to paradox. For assuming that the evidence of section 3 indicates that X-PPs are not subjects clashes with the extraction evidence, which then seemingly becomes a web of unexplained anomalous correlations. The paradox then is that there appears to be no acceptable conclusion as to the subject status of the X-PPs of NEX clauses. Since, as indicated in (1), this is a theoretically intolerable situation, it must be due not merely to the sort of factual observations that underlie sections 2 and 3 but also to some false assumption(s) that have colored their interpretation. It must be assumed that if one can find and reject these incorrect premises, a consistent treatment of all the facts will emerge. Before focusing directly on that step, I consider further data that greatly strengthens the view that X-PPs are not S3s and thus suggests that the proper area to search for the false and paradox-inducing assumptions lies in the previously assumed interpretation of the extraction patterns of section 2.

5. Null expletive analyses of NEX clauses 5.1. Criticisms of null expletive analyses Postal (1977) proposed that the S3 of NEX clauses was an invisible expletive NP, specifically an invisible variant of presentational there; that is, a version of (4a) was advocated. Although Bresnan herself notes (1994: 99) that an expletive subject analysis explains various facts that are problematic for her own X-PP as subject treatment, facts that include agreement determination, she (1994: 98–103) criticizes expletive analyses and ultimately concludes that they are inferior to her alternative. One can divide her criticisms into two categories. The first simply stresses that (previous) expletive subject analyses offer no basis for the facts of section 2, that is, in general, for the fact that extracted X-PPs obey extraction constraints otherwise apparently exclusively linked to subjects. This category of criticism is surely sound as far as it goes; no extant expletive analysis has offered any basis for the fact that X-PPs obey the extraction constraints otherwise linked to subject extraction. The second category consists of all Bresnan’s other criticisms, and these are, I will argue, largely without force. Consideration of them ultimately strengthens the case made in section 3 that X-PPs ae not subjects. 5.2. Lack of expletive ‘sources’ Bresnan (1994: 99) claims that an obvious problem for the expletive subject view is that certain types of locative inversion NEX cases, particularly those that involve

30


directionals, cannot occur with overt expletive subjects. At issue are contrasts like the following: (44)

Bresnan (1994: 99 [76]) a. Into the room (*there) ran Mother. b. Out of it (*there) steps Archie Campbell. c. About a half an hour later in (*there) walk these two guys. d. Home (*there) came John.

Similar data were noted earlier by Green (1985: 125-126).14 Bresnan (1994: 100) concludes from these data that many NEX clauses “lack a plausible source.” However, the term ‘source’ here is neither defined nor explicated. This term has (or had) a standard usage in transformational work. But Bresnan herself has long since abandoned such frameworks. Moreover, even internal to those ideas, at least as widely interpreted, contrasts like those in (44) would hardly impose the view that the shorter versions could not be related to structures that contain invisible variants of the parenthesized elements. For instance, while it may be false that (45a) has a ‘source’ in which the fronted phrase is in postverbal position, surely (45b) does not suffice to show that: (45)

a. No matter what Ed does turns out badly. b. *Ed does < (no matter) what > turns out badly.

In short, Bresnan has failed to spell out the logic of the supposed criticism of the expletive analysis from facts like (44) and her appeal to a vague term like ‘source’ makes it seem like there is some straightforward basis for the conclusion when there is not. An advocate of the expletive analysis is free to say about (44) that nothing more is at issue than the conditions under which the expletive must (as opposed to can) be invisible. Bresnan herself (1994: 99) notes that many locative inversion cases have alternates that contain a visible there. An even stronger connection was noted much earlier: (46)

Green (1985: 121–122) “For the most part, V-inversions correspond to there-constructions, with or without an initial non-subject phrase. That is, for almost every inversion sentence of the form of (16a), there is a there-sentence of the form (16b) or (16c), or both, where C is in (A, P, V): 16a. CP . . . V NP W 16b. CP there . . . V NP W 16c. There . . . V NP CP W”

I return to such correlations later. Thus an overall grammar of English must specify the conditions of alternation in some way. Unless Bresnan can argue that the expletive hypothesis about NEX clauses complicates this task or renders it impossible, which she has not attempted, the conclusion that (44) infirms a null expletive analysis borders on the non sequitur.


31

A similar remark holds for the point that Bresnan recalls from Aissen (1975) to the effect that the definiteness constraint on the P-subject that occurs with expletive there is not really found in NEX cases: (47)

a. b. c. d.

*In the closet there still sat Fido. In the closet still sat Fido. *Suddenly there ran out of the woods Bob and Louise. Out of the woods suddenly ran Bob and Louise.

But again there is no argument for the conclusion that contrasts like (47) are problematic for the null expletive view of NEX clauses (“show that [47b and d] lack sources” in Bresnan’s terms). For the advocate of such an analysis can say that the definiteness constraint, highly mysterious and poorly understood despite a great deal of work (see, e.g., Reuland and ter Meulen, 1987; and Safir, 1985), is stated in such a way as to be sensitive in part to the difference between visible and invisible expletives, more plausibly, perhaps, to other structural factors that determine that. Note, too, that the sensitivity is partial, in that there are many fine there cases with definite DPs; see, for example, Rando and Napoli (1978) and examples like (48): (48)

a. Into that cave (there) just crawled the largest snake I have ever seen. b. At that time (there) was said to have been rounded up the entire membership of the Communist Party. c. Under those conditions (there) will certainly still remain the problem of repatriating the refugees.

5.3. Typological implausibility 5.3.1. The logic of the case Another in principle very weak sort of objection to the null expletive hypothesis that Bresnan makes is that such an analysis is typologically implausible. This is based on facts about other languages that allow null subjects in a range of positions where English does not. The point involves the claim that “[n]o other properties of English point to the presence of such null ‘pro’ subjects” (1994: 102). Even if this assertion were true, the argument against a null expletive analysis would be weak. An advocate of such could simply claim that not enough is known about NL grammar to ground any principle that would exclude an NL with the following two properties: (i) it has no finite clauses with null subjects that are not parallel to NEX clauses and (ii) it has clauses parallel to NEX clauses that have null expletive subjects. In the absence of such principles, Bresnan’s rejection of the null expletive analysis on typological grounds has little weight. 5.3.2. The facts of the case Moreover, a much stronger defense against the typological criticism is available, since the assumption that English has no null subject finite clauses outside the realm of NEX clauses, though no doubt accepted by others beside Bresnan, is arguably false.

32


I suggest that there are several types of such clauses and, ironically, Bresnan’s thesis (1972: 136) already has discussed one of them. Let us examine first what are called as-Parentheticals (A-Parentheticals) in Postal (1994b). These are illustrated in (49): (49)

a. Lasers can, as he has long felt, cut through stone walls. b. Lasers can, as he feared, cut through animate tissue. c. Lasers cannot, as it first appeared, cut through stone walls.

The instances of as in these forms roughly correspond to the that clauses of independent clauses. Note, for instance, that as in (49c), A-Parentheticals are formed from expressions that occur essentially only with that clauses: (50)

a. It appeared that lasers could cut through that wall. b. *It appeared a vision/something/many colors/those possibilities.

Consider then cases like (51): (51)

a. Lasers can, as is obvious, cut through stone walls. b. Lasers can, as was proved by Mike, cut through stone walls.

With adjectival and passive predicates, A-Parentheticals occur without an obvious candidate for S3 status. Let us for convenience refer to these cases as M(ysterious)A-Parentheticals. There are at least three reasonably straightforward proposals about their structures: (52)

a. The as in M-A-Parentheticals is an extracted S3. b. The as in M-A-Parentheticals is an extracted non-S3 and there is a null (expletive) S3. c. The as in M-A-Parentheticals is indifferently an extracted S3 or non-S3.

Incidentally, nothing relevant here is changed if (as in Potts, 2002), instead of regarding as as an extractee, it is taken to be some grammatical element, for example, a P, that is cooccuring with an invisible extracted phrase. I suspect that the common view would be (52a); see Kayne (1984: 67) and Tellier (1988: 134–135). This is pretty much the obligatory choice for anyone who adopts Bresnan’s (1994) position, since either (52b or c) recognizes that English has at least some null subject finite clauses. But there is evidence that (52b) is correct and that cases like (51a and b) involve invisible extraposition expletives of the type that otherwise show up as it, as in (49c) and (53b): (53)

a. It seemed to everyone that lasers could cut through stone walls. b. Lasers cannot, as it had previously seemed to everyone, cut through stone walls.

In (53b), the as clearly corresponds to a nonsubject, given that the that clause complement of a verb like seem can never appear as its subject:


(54)

33

*That lasers can cut through stone walls had seemed to everyone.

Thus a manifest property of English syntax is: (55)

A-Parentheticals can be formed from nonsubjects.

Important evidence that bears on the status of (55) derives from a class of complement-taking verbs discussed in Chomsky (1981: 122), Marantz (1984: 133), Postal (1986a: 96–99; 1998: chapter 4, note 12); Williams (1981); and see chapter 8, section 8. These verbs do not permit their complements to appear as their subjects (in passives). Members of this class include feel, hold, say, suppose, and think: (56)

a. Everyone intelligent feels/holds/says/supposes/thinks that gold is rare. b. *That gold is rare is felt/held/said/supposed/thought by everyone intelligent. c. It is felt/held/said/supposed/thought by everyone intelligent that gold is rare.

Nonetheless, these verbs permit passive M-A-Parentheticals just like complementtaking verbs such as believe, which are not subject to the constraint blocking (56b): (57)

Gold is not, as is deeply felt/widely held/sometimes said/usually supposed/generally thought, extremely rare.

The existence of M-A-Parentheticals like (57) combines with the facts in (56) to support much more than (55). Such cases argue that the as in passive M-A-Parentheticals does not correspond to structures like (56b), that is, those with passive that clause subjects, but rather to those like (56c), that is, those with (extraposition) expletive subjects (S3s). If so, (52a) is wrong and (52b) arguably the correct position. The point is strengthened because the verbs in (56) also do not permit objectraising versions of transitive clauses with complement objects. Nonetheless, M-AParentheticals that correspond to those are possible: (58)

a. That Herb is a complete hypocrite is easy to believe/*feel/*hold/*say/*suppose/ *think. b. Herb was not, as was (nonetheless) easy to believe/?feel/hold/say/suppose/think, a complete hypocrite. c. It is easy to believe/feel/hold/say/suppose/think that Herb is a complete hypocrite.

These data also argue that gaps in M-A-Parentheticals like those of (58b) correspond to nonsubjects, here to structures like those of (58c). Further support for this view emerges as follows. The facts of (56)–(58) are consistent with a much stronger position than (55), namely: (59)

A-Parentheticals can only be formed from nonsubjects.

I suggest that (59) is true, which excludes analysis (52a) independently of facts like (56)–(58). One sort of evidence for (59) could take the form of complement-taking

34


verbs that do not, in simple clauses, permit their complements to be nonsubjects but do permit them as subjects. According to (59), such verbs could not permit (any) A-Parentheticals—in particular, no passive M-A-Parentheticals. As observed in Grimshaw (1982) and discussed in Dowty and Jacobson (1988: 103), Hukari and Levine (1991: 116–117), Jacobson (1992b), and Postal (1998: 108–114), such a class of verbs (see chapter 8, section 8) is represented by capture, express, and reflect: (60)

a. *This theory captures/expresses/reflects that languages have verbs. b. *It is captured/expressed/reflected by this theory that languages have verbs. c. That languages have verbs is captured/expressed/reflected by this theory.

Given these facts, (59) predicts, correctly, that these verbs preclude A-Parentheticals:15 (61)

a. *Languages do (not) have, as this theory captures/expresses/reflects, the sort of verb in question. b. *Languages do (not) have, as is captured/expressed/reflected by this theory, the sort of verb in question.16

Further evidence with the same essential logic derives from data like (62): (62)

a. That she is greedy strikes me as the problem. b. *It strikes me as the problem that she is greedy. c. *She is greedy, as strikes me as the problem.

That is, (62a) indicates that a subject that clause is possible, while (62b) shows that extraposition of the subject is not. Hence the complement can only appear as subject. And (62c) documents that a corresponding M-A-Parenthetical is impossible, as (59) determines. Principle (59) receives independent support from the fact that it can be taken as a special case of a more general principle that covers not only A-Parentheticals but also those called N(ull)-Parentheticals in Postal (1994b). These are illustrated in (63): (63)

a. Jerome is, I believe, quite intelligent. b. Jerome is, it is believed, quite intelligent. c. Jerome is, it is obvious, quite intelligent.

The relevant point is that the gap in an N-Parenthetical can never correspond to a subject: (64)

a. *Jerome is, is believed, quite intelligent. b. *Jerome is, is obvious, quite intelligent. c. *Jerome is, appears to me, quite intelligent.

Some principle must thus guarantee this characteristic, and a priori (65a) is preferable to the less general (65b):


(65)

35

a. English parentheticals can only be formed from nonsubjects. b. English N-Parentheticals can only be formed from nonsubjects.

But (65a) trivially entails (59), which is then arguably just a special case of a broad generalization about English parenthetical formation. So far then, triple evidence has been amassed for the view that passive M-AParentheticals involve an invisible subject that corresponds to the extraposition expletive and not to the that clause or as—in other words, that in these cases as does not represent an extracted subject. First, in (57) and (58) M-A-Parentheticals are good, although the verbs in question bar subject that clauses, while allowing extraposed ones. Second, in (61) and (62) M-A-Parentheticals are bad, although the verbs in question allow subject that clauses, while banning nonsubject ones, including extraposed ones. Thus the formation of A-Parentheticals patterns with extraposed that clauses, not with subject that clauses. Third, it is independently supported by the facts of N-Parentheticals that English parentheticals cannot involve subject extraction. The view consistent with this evidence is (52b), while (52a) is inconsistent with it. Despite the evidence just referred to, a defender of Bresnan’s position that there are no null (expletive) subject clauses in English might still try to maintain (52a) in some way. However, beyond what has been argued so far, this is in effect impossible internal to Bresnan’s overall set of assumptions about NEX clauses, given the facts illustrated in (66): (66)

a. Such pistols are, as was obvious and Frank just demonstrated, quite deadly. b. Such pistols are, as Frank just demonstrated and was obvious anyway, quite deadly c. Such pistols are, as has long been suspected and Frank just demonstrated, quite deadly.

I find all of these perfect. But, according to (52a), the first conjuncts of (66a and c) each involve an extracted subject, while the second involves an extracted nonsubject, both at the highest level of coordination. That analysis violates the Parallelism Constraint, one of Bresnan’s basic arguments for the X-PP as subject view of NEX structures, discussed in section 2.3. Example (66c) shows that the order of conjuncts is irrelevant. On the contrary, under the null (extraposition) expletive analysis of (66) advocated here, consistency with the Parallelism Constraint is maintained, as extraction from both conjuncts involves nonsubjects in all cases. The evidence in (66) is strengthened by a contrast between these as facts and partly parallel facts that involve sentential which, often contrasted with as in other ways; see Postal (1994b), Potts (2002), Rizzi (1990: 15), and Ross (1984): (67)

a. *He said cyanide pistols are quite deadly, which is obvious and I have always believed. b. *He said cyanide pistols are quite deadly, which I have always believed and is obvious. c. *He said cyanide pistols are quite deadly, which was long suspected and Frank just proved. d. *He said cyanide pistols are quite deadly, which Frank just proved and was long suspected.

36


The (66)/*(67) contrast will follow from the assumption that the former correspond to constituents extracted from extraposed positions, while the latter require extraction from standard DP positions. It is easily seen that, independently, sentential which cannot be formed from an extraposed position: (68)

a. *That he was sick seemed to everyone. b. It seemed to everyone that he was sick. c. *He was sick, which it had seemed to everyone.

Overall then, an additional argument for (52b) and against (52a), hence against the claim that English does not allow null expletive clauses, is that such a view is incompatible with the Parallelism Constraint. The status of a view of M-A-Parentheticals that involve (52a) and is consistent with Bresnan’s overall assumptions about English clause structure in general and NEX cases in particular is even worse than so far indicated, as shown by (69): (69)

a. Ted was cheated, as I assumed (*that) was obvious. b. Ted was cheated, as I thought (*that) had been proved by Michelle.

Examples (69a and b) show that the very M-A-Parentheticals that the distributional evidence from verb/that clause restrictions shows involve constituents extracted from extraposed positions, not from subject positions, and would violate the Parallelism Constraint if they did involve extracted subjects nonetheless obey the Anticomplementizer Constraint. Although not directly about NEX cases, this fact is, I take it, ultimately devastating for Bresnan’s approach to NEX cases. For it indicates that a primary assumption that underlies her conclusion that X-PPs are subjects, the view that both the Anticomplementizer Constraint and the Parallelism Constraint are defined on subjects (S3s), is just false. The interaction between A-Parentheticals and these constraints combined with the multiple grounds for denying that M-A-Parentheticals involve subject extraction shows this cannot be the case. 5.3.3. Comparative cases The evidence cited so far in section 5 that involves A-Parentheticals can essentially be duplicated with comparative structures. Key data is already found in Bresnan (1972: 136), which supplied the following paradigms, crediting in part William Leben: (70)

a. b. c. d. e.

More is known than it seems. *More is known than __ seems. *More is known than it is necessary. More is known than __ is necessary. More is known than __ seems to be necessary.

A key point for our purposes here is that these patterns match significantly those of A-Parentheticals:


(71)

a. b. c. d. e.

37

He knew that, as it seemed. *He knew that, as seemed. *He did that, as it was necessary. He did that, as was necessary. He did that, as seemed to be necessary.

Examples like (71a) show that with comparatives English grammar permits the nonappearance of any remnant of a that clause complement. As with A-Parentheticals, the question then arises whether this phenomenon of complement vanishing is possible corresponding to subject complements. Evidence to the negative is that A-Parenthetical/comparative parallels extend to the unique classes of verbs that support the claims that A-Parentheticals preclude subject extraction, that is, that support the claim that M-A-Parentheticals involve null expletive subjects: (72)

a. More is known than (*it) is generally felt/held/said/supposed/thought. b. More is known than is explained/*captured/*expressed/*reflected by this theory.

All these facts support the view that like both types of parentheticals, vanished complement comparatives cannot correspond to subject complements. In short, there is every reason to think that comparative examples like (71d), which parallel M-A-Paentheticals, do not involve subject extraction but rather null expletive subjects. And like A-Parentheticals, these cases then show that the Anticomplementizer Constraint is not uniquely linked to subject extraction, since such effects manifest in comparative cases of this kind: (73)

a. More is known than he believes (*that) is necessary. b. More is known than he said (*that) had been reported.

5.4. Further evidence for a null expletive analysis of M-A-Parentheticals 5.4.1. Remarks Beyond the arguments in section 5.3, a number of other rather clear grounds clash with the view that M-A-Parentheticals involve extracted S3s. These also show, rather, that the as in such cases corresponds to an extraposed clause and that the S3 is an invisible extraposition expletive. 5.4.2. Number agreement Conjoined subject that clauses can determine plural verb agreement, which is impossible with the extraposition expletive: (74)

a. That the company is bankrupt and that he is responsible are/is obvious. b. It is/*are obvious that the company is bankrupt and that he is responsible.

38


The M-A-Parenthetical facts parallel the extraposition cases and contrast with the subject ones: (75)

The company is bankrupt and he is responsible, as is/*are (both) obvious.

5.4.3. Standard floating quantifiers Conjoined subject that clauses can link to SFQs; extraposed clauses cannot: (76)

a. That the company is bankrupt and that he is responsible are both obvious. b. *It is/are both obvious that the company is bankrupt and that he is responsible.

Again, the facts for M-A-Parentheticals match those of the extraposed cases: (77)

*The company is bankrupt and he is responsible, as is/are both obvious.

5.4.4. Floating emphatic reflexives Subject that clauses can link to FERs; extraposed clauses cannot: (78)

a. That the company is bankrupt and that he is responsible are themselves obvious. b. *It is/are itself/themselves obvious that the company is bankrupt and that he is responsible.

Once more, M-A-Parentheticals match extraposed structures: (79)

*The company is bankrupt and he is responsible, as is/are itself/themselves obvious.

5.4.5. Summary The facts of plual agreement, SFQs, and FERs further confirm the conclusion that M-A-Parentheticals do not involve extracted S3s but rather represent instances of an extracted constituent that corresponds to an extraposed clause, with a null extraposition expletive functioning as S3.17 5.5. Results The implications of this brief look at A-Parentheticals and subjectless comparatives appear to be very strong. First, English arguably has null expletive clauses independently of the analysis of NEX structures. So any attempt to exclude a null expletive analysis on grounds of plausibility, ‘typological’ or otherwise, collapses. More strongly still, the assumption that the Anticomplementizer Constraint and the Parallelism Constraint are both defined on subjects (S3s), which is absolutely central to Bresnan’s conclusion that the X-PPs of NEX structures are subjects, is independently untenable for English and thus cannot be appealed to in the analysis of NEX cases.


39

6. The NEX clause paradox partially resolved The conclusions of section 5 suggest, I believe, a resolution of the paradox that involves the subject status of X-PPs in NEX clauses. One term of the paradox arose principally from Bresnan’s striking apparent demonstration that extracted X-PPs apparently alone among extracted elements that are not obviously subjects behave like extracted subjects. This led Bresnan to the conclusion that X-PPs are subjects. The other term arises from the fact, argued in section 3, that in a wide variety of other ways X-PPs behave like nonsubjects. A solution has become visible based on the conclusion that the extraction evidence does not show that X-PPs are subjects because it is not true that the Anticomplementizer Constraint and the Parallelism Constraint are both defined on subjects (S3s). Rather, the A-Parenthetical facts show clearly that the former is not defined (exclusively) over S3 extraction. This evidence leaves how the latter is defined open. But overall, the new situation now makes it reasonable to see the problem in terms entirely different from Bresnan’s. The question is not whether X-PPs are subjects; they clearly are not. The question is how the extraction constraints are to be stated in such a way that subjects and X-PPs but not other arbitrary constituents are subject to them. This formulation is not even quite correct. As section 5 indicated, it is also necessary to subsume nonsubject extraction of certain as and comparative cases under the Anticomplementizer Effect. I return to the formulation of the extraction constraints in section 9. First, I consider additional facts about NEX cases that both contribute further to undermining Bresnan’s assumptions and further support the view that these involve null expletive subjects.

7. Further evidence for invisible expletives in NEX clauses 7.1. Remarks Further grounds for analyzing NEX clauses as having null expletive subjects can take the form of direct evidence for the presence of invisible expletives. This will involve chiefly respects in which NEX clauses share properties with clauses that contain visible expletive there. A good way to begin is by consideration of some further arguments of Bresnan (1994) against the presence of null expletives, which I will argue do not show anything of relevance. 7.2. Two arguments from Hoekstra and Mulder Section 5 already considered Bresnan’s definiteness and ‘typological plausibility’ arguments. Bresnan (1994: 102) considers two further arguments against null expletives, due in effect to Hoekstra and Mulder (1990). The first is based on examples like (80): (80)

a. We all witnessed how down the hill came rolling a huge baby carriage. b. We suddenly saw how into the pond jumped thousands of frogs.

40


These are supposed to show that it is not possible to capture the fact that NEX clauses manifest the Anticomplementizer Constraint by claiming that the null expletive subject is “licensed only by a locative PP in the same position as WH-extracted phrases.” I do not really follow how this argument yields an objection to null expletives; it is supposed to depend on the idea that cases like (80) show that the position of extracted WH-phrases (SPEC, CP) is already filled, by the how. But I believe examples like (80) are irrelevant, since the best analysis of this type of how is probably that it is a restricted complementizer, thus parallel to that and not to a WH-phrase. It is surely not a question form of a manner adverbial. Observe such contrasts as the following: (81)

a. They may wonder how else we could proceed. b. Did they witness how (*else) down the hill came rolling a baby carriage?

(82)

a. They may ask how on earth a bright light could have shone into the room. b. They did not testify how (*on earth) into the room suddenly shone a bright light.

In any event, I see no clear way in which these facts bear on the posit of a null expletive in NEX clauses. The second argument is more straightforward: (83)

Bresnan (1994: 102) “questioning the inverted locative fails to trigger auxiliary inversion (I-to-C movement), which is obligatory where there is a subject and impossible where the subject itself has been extracted.”

That is, a paradigm like (84) parallels a standard subject one like (85), where both contain unstressed auxiliaries: (84)

a. On which wall hung a portrait of the artist? b. *On which wall did hang a portrait of the artist?

(85)

a. Which wall contained a portrait of the artist? b. *Which wall did contain a portrait of the artist?

That the locative behaves likes an extracted S3 argues that it is such, incompatible with the view that there is a null expletive subject. While this argument from unstressed auxiliaries is sound as far as it goes, it doesn’t go far enough. For consider question S3 extraction when the auxiliary do is stressed: (86)

Which wall (must we conclude) DID contain a portrait of the artist?

In this case, a (stressed) auxiliary is possible in the presence of an extracted subject. If then the X-PP of a NEX clause were an S3, one would expect it to be compatible in the question case with a stressed auxiliary do. But this is impossible:


(87)

41

a. On which wall HUNG a portrait of the artist? b. *On which wall (must we conclude) DID hang a portrait of the artist?

In short, the two auxiliary arguments seem to cancel each other out. While the X-PP resembles an S3 in weak auxiliary behavior, its strong auxiliary behavior resembles that of a non-S3. No conclusion about the subject status of X-PPs can thus so far be drawn from auxiliary presence or absence data. 7.3. Presentational there properties reflected in NEX clauses 7.3.1. Setting My position is that NEX clauses are a variety of presentational there structure in which, for reasons returned to in section 9, the expletive is invisible. Evidence for this view involves properties that NEX clauses share with presentational there structures. The recognition of such sharing is hardly new; recall Green (1985). And Coopmans (1989: 743) asserts: “Many of the restrictions on stylistic inversion in the context of preposed adverbial PPs also hold for the inversion in the presentational there construction . . .” 7.3.2. P-subject properties As Bresnan (1994: 99, 107) herself notes, the postverbal NPs of presentational there clauses determine verb agreement, so a null expletive account of NEX clauses reduces the agreement properties of these, really anomalous under Bresnan’s account, to the independently needed principles operative in presentational-there structures: (88)

On the rock (there) *was/were sitting two giant frogs.

Second, the postverbal NPs of presentational there clauses cannot be definite anaphoric pronouns (see Bresnan, 1994: 86 for recognition of the restriction in NEX cases), isolated indefinite pronouns like someone, or single-word wh pronouns.18 NEX structures share all these features: (89)

a. *On the rock (there) was sitting him/someone. b. *Who said that on the rock (there) was sitting who?

Third, consider tags in yes/no questions. As Bresnan (1991) observes, in tag questions a declarativelike clause is followed by a tag that consists of an auxiliary verb and a pronoun that agrees with the S3 of the declarative constituent: (90)

a. Guns are dangerous, aren’t they/*it/*you? b. Selma will cry, won’t she/*they/*us?

Bresnan then cites an observation from Bowers (1976: 237) that the latter took to argue against the S3 status of the P-subject of NEX clauses. This is the grammaticality of tag question variants of NEX cases like (91), in which the tag pronoun is there:

42 (91)


a. In the garden is a beautiful statue, isn’t there? b. In the garden stood/lay a beautiful statue, didn’t there?

Bresnan (1991) asserts that “the hypothesis that the inverted locative is the topicalized subject would explain this situation.” She is thus claiming that the choice of the tag pronoun there supports hypothesis (4c). But these facts of course fail to support (4c) over the view that NEX clauses involve an invisible expletive otherwise identical to there. Moreover, when one considers nonlocative NEX cases, Bresnan’s claim breaks down in that there are topicalized phrases in such cases that would not determine there in anaphoric contexts and yet whose best tag must contain there: (92)

a. That task fell to Gloria1 but it shouldn’t have fallen to her1/*there1. b. To Gloria will fall a number of unpleasant tasks, won’t *her/?there? c. They built a number of warships [at that time]1, but they didn’t deploy them then1/ *there1. d. At that time were built a number of warships, weren’t *then/there? e. The following answer corresponds to [that question]1, but he didn’t ask it1/*there1. f. To that question might correspond an interesting answer, mightn’t *it/?there?

Facts like those in (92) argue that the matching between the tag pronoun and the NEX clause involves an expletive there and not the topicalized PP. If so, the tag facts do not, contrary to Bresnan’s claim, support position (4c) but actually contribute to undermining it.19 7.3.3. Context parallels 7.3.3.1.   Presentational there clauses and NEX clauses are subject to much the same sort of overall global constraints on the predicate classes that permit them. First, neither type of structure can be formed with intransitive predicates that take predicate nominals:20 (93)

a. Many women are lawyers in New Zealand. b. *In New Zealand (there) are lawyers many women.

(94)

a. Certain unfortunate individuals turn into werewolves after sundown. b. *After sundown (there) turn into werewolves certain unfortunate individuals.

7.3.3.2.   Second, neither presentational there clauses nor NEX clauses can be based on adjectival predicates:21 (95)

a. That sort of heroin addict is prone to accidents on the highways. b. *On the highways (there) is prone to accidents that sort of heroin addict.

(96)

a. The majority of young people are hesitant about that in Mexico. b. *In Mexico (there) are hesitant about that the majority of young people.


43

A corollary of the ban on adjectival NEX clauses is that there are no NEX versions of adjectival passives, a type of structure discussed in, for example, Bresnan (1982c), Dryer (1985), Levin and Rappaport (1986), Siegel (1973), Wasow (1977), and Williams (1981):22 (97)

a. b. c. d.

Regular Passive: Several boxes were loaded with cookies. Adjectival Passive: Several boxes remained loaded with cookies. In that room (there) were loaded with cookies several large, square metal boxes. *In that room (there) remained loaded with cookies several large, square metal boxes.

(98)

a. b. c. d.

Regular Passive: Several bottles of wine were being tasted in that kitchen. Adjectival Passive: In that kitchen several bottles of wine went untasted. In that kitchen (there) were being tasted several bottles of rare wine. *In that kitchen (there) went untasted several bottles of rare wine.

7.3.3.3.   A significant fact about NEX structures noted in Bresnan (1995) is that in no case can the P-subject be a complement clause, that is, a non-WH infinitive or that clause:23 (99)

Bresnan (1995) a. (The warning) that enemies were coming was written on the roof. b. On the roof was written the warning that enemies were coming. c. *On the roof was written that enemies were coming.

(100)

a. b. c. d.

The task of calling Sam fell to Mary. It fell to Mary to call Sam. To Mary fell the task of calling Sam. *To Mary fell to call Sam.

(101)

a. From that testimony can be deduced the fact that he is a thief. b. *From that testimony can be deduced that he is a thief.

(102)

a. At that conference was suggested a new line of research. b. *At that conference was suggested that we abandon the project.

Notably, though, identical restrictions hold for clauses with explicit presentational there: (103)

a. On the roof there was written the warning that enemies were coming. b. *On the roof there was written that enemies were coming.

(104)

a. To Mary there fell the task of calling Sam. b. *To Mary there fell to call Sam.

(105)

a. From that testimony there can be deduced the fact that he is a thief. b. *From that testimony there can be deduced that he is a thief.

44


(106)

a. Only at that conference was there suggested a new line of research. b. *Only at that conference was there suggested that we abandon the project.

One might state these facts as simply indicating uniformly that the extraposition expletive is it, not there. 7.3.3.4.   Third, despite earlier examples like (7c), in which NEX structures can be formed on raising to object structures, there are various partly parallel environments where both there and NEX clauses are impossible: (107)

a. Ellen prevented there from being a riot. b. *There was prevented from being a riot.

The generalization of relevance about this environment is that expletive there is unpassivizable in it. Most other NPs, even expletive ones, are not:24 (108)

a. Students were prevented from rioting. b. It was prevented from raining (by Zeus). c. It must be prevented from becoming obvious that we are spies.

Notably then, presentational there and NEX clauses both seem to share this constraint: (109)

a. *Ellen prevented there from dashing into the meeting three angry students. b. *Into the meeting (*there) were prevented from dashing three angry students.

Of course, the generalization specified does not account for *(109a); there may be a more general constraint that bans even unpassivized expletive there of the presentational type. If so, NEX clauses share this constraint as well, as (110) indicates: (110)

*Into the room he prevented from dashing three angry students.

Compare: (111)

a. Into the meeting she believed (there) to have dashed three angry students. b. Into the meeting (there) were believed to have dashed three angry students.

7.3.3.5. ⁄ ⁄    ment in (112): (112)

a. b. c. d. e.

Next consider the environ-

You can count/depend/rely on her to tell the truth. You can count/depend/rely on it to rain during the picnic. You can count/depend/rely on it to be unclear whether he is guilty. *You can count/depend/rely on there to be a riot. *You can count/depend/rely on there to have arrived some penguins by that time.


45

The generalization appears to be that, with these verbs, the object of on with a following infinitive cannot, for the most part, be expletive there, although other expletives are permitted. Limitations on this constraint are discussed in chapter 2. The relevant constraint extends to pseudopassives of such cases: (113)

a. It can be counted/depended/relied on to rain/to be obvious that he is guilty. b. *There can be counted/depended/relied on to be a riot/to have arrived some penguins by that time.

Strikingly, both active and passive versions of NEX structures obey the constraint as well: (114)

a. *On the bench you can count/depend/rely on to be sitting two penguins. b. *By that time can be counted/depended/relied on to have arrived some penguins.

7.3.3.6.    Next, consider perception verb contexts like the following: (115)

a. I sometimes heard/saw/watched her being hassled in Jane’s class. b. In Jane’s class, I sometimes heard/saw/watched her being hassled.

Notably, it is not possible to form NEX structures on this model: (116)

*In Jane’s class, I sometimes heard/saw/watched being hassled the kid you are talking about.

This parallels (117): (117)

*In Jane’s class, I sometimes heard/saw/watched there being hassled the kid you are talking about.

Actually, the claim about NEX structures and perception verbs is too general. There are cases where the two can combine. But, notably, in those cases, the variant with there is also found: (118)

a. In Portugal, I saw a new religion arise. b. In Portugal, I saw (there) arise a new religion. c. In Portugal, I saw (there) develop a great revulsion toward foreigners.

So again the distribution of NEX clauses closely matches that of there structures. 7.3.3.7.     Another environment that links NEX clauses to presentational there ones was touched on in section 3; see (18):

46


(119)

a. b. c. d. e. f.

It looks/sounds like it is impossible to do that. It looks/sounds like it will rain. There looks/sounds like there is going to be a riot. There looked/sounded like there were two cats fighting in the yard. *There looks like there fell to Ruth the task of cleaning the pots. *There sounds like there were fighting two wild boars.

Here the constraint appears to bar the expletive there of (only) presentational there structures, including that of P-there structures. But NEX forms are also barred: (120)

a. *To Ruth looks/sounds like (there) fell the task of cleaning the pots. b. *In the woods sounds like (there) are fighting two wild boars.

7.3.3.8.  Another link between NEX clauses and there structures is that both are incompatible with middle formation, which contrasts in this respect with participial passives:25 (121)

a. In this institution, inmates bore easily. b. *In this institution (there) bore easily many inmates.

(122)

a. b. c. d.

At home, many such new products prepare quite rapidly. At home, many such new products can be prepared quite rapidly. *At home, (there) prepare quite rapidly many such new products. At home, (there) can now be prepared rapidly many such new products.

(123)

a. b. c. d.

In that state, all multiple murderers used to hang. In that state, all multiple murderers used to be hung. *In that state, (there) used to hang all multiple murderers. In that state, (there) used to be hung many multiple murderers each year.

7.3.3.9.   English participial passives appear in at least two forms, with main verb be or get: (124)

a. In that field dozens of partisans were/got executed. b. In that way several new recruits were/got trained.

But neither presentational there nor NEX structures are possible with the get varieties: (125)

a. In that field (there) were/*got executed dozens of partisans. b. In that way (there) were/*got quickly trained a number of new recruits.

7.3.3.10.       Bresnan (1994: 79–80) observes insightfully that although passives are a very productive source of well-formed NEX cases, not all passives permit NEX variants. In particular, she claims that pseudopassives (prepositional passives) never allow this:


(126)

47

Bresnan (1994: 79) a. We fought for these rights in these very halls. b. These rights were fought for in these very halls. c. *In these halls were fought for these rights. d. In these halls were fought tremendous battles for equal rights.

While I am not entirely sure that Bresnan’s claim holds for the full class of NEX passives, I believe that whenever a NEX variant of a pseudopassive is barred, a parallel constraint holds when an explicit there is present: (127)

a. In the hall, some touchy issues about women’s rights were being considered/discussed/debated/argued about/talked about/fought about (by the delegates). b. In the hall, (there) were being considered/discussed/debated/*argued about/ *talked about/*fought about in a raucous way some touchy issues about women’s rights.

See also: (128)

a. In that hospital, many victims of the explosion are being treated/taken care of/ worked on/looked after by only three doctors. b. In that hospital, (there) are being treated/*taken care of/*worked on/*looked after many of the victims of the explosion.

Next, consider passives based on double object verb constructions. Postal (1986a) characterized the three types of passive possible with certain double object verbs as in (129): (129)

a. b. c. d. e.

Marsha gave Louise the books. Marsha gave the books to Louise. Primary Passive The books were given to Louise. Secondary Passive Louise was given the books. Tertiary Passive The books were given her/Louise.

While many current speakers do not accept any tertiary passives, many, including the present writer, do; see chapter 8. Bresnan (1994: 79, note 9) observes in effect that NEX structures cannot be formed on secondary passives, which she ultimately attributes to a putative basic constraint on NEX structures that the ‘shifted’ element (my P-subject) must be a thematic direct object:26 (130)

a. In that hall were given to Louise insightful books on wart prevention. b. *In that hall were given them yesterday several recent graduates.

Bresnan did not observe though that even for speakers who accept tertiary passives, there are no NEX variants of the latter, even though these would meet all of the conditions on the formation of such she adduces. Notably then, parallel constraints hold for presentational there structures:

48


(131)

a. In that hall (there) were given to Louise insightful books on wart prevention. b. *In that hall (there) were given Louise/her insightful books on wart prevention.

Some might argue that cases like (131b) violate the constraint that NEX structures are not possible in the presence of a direct object. However, it can be argued, I believe, that, for example, Louise in (131b) is an indirect object, not a direct object, so that no violation of a direct object constraint could exist. There are other cases, typically nonagentive, where passives do not permit NEX variants, for currently obscure reasons. These also seem to systematically preclude variants with explicit there. Compare: (132)

a. At that time (there) were reached/*hated certain conclusions. b. During the period in question (there) were discovered/*forgotten several important theorems.

7.3.3.11.   It has been known since Cantrall (1969: 124) (see Aoun, 1985: 81–83; Dresher and Hornstein, 1979; Postal, 1974: 198–204) that there are constraints on multiple raisings of certain types of DPs, including expletive there. Compare: (133)

a. Jerome continues to seem to support the director. b. It continues to seem to be impossible to find her. c. *There continue to seem to be riots in Bananastan.

(134)

a. The captain appeared to start to be interested in Greta. b. It appeared to start to be possible to treat that disease. c. *There appeared to start to be riots in Bananastan.

(135)

a. Clara is believed to continue to undergo therapy. b. It is believed to continue to be impossible to square a circle. c. *There are believed to continue to be riots in Bananastan.

Where earlier evidence seems to have always involved existential there, such constraints also hold for presentational there: (136)

a. b. c. d.

Into that cave there crawled strange-looking frogs. Into that cave there continue to crawl strange-looking frogs. *Into that cave there continue to seem to crawl strange-looking frogs. *Into that cave there are believed to continue to crawl strange-looking frogs.

Once again, the constraint on presentational there is mirrored in constraints on NEX structures, which also resist multiple raising: (137)

a. Into that cave crawled strange-looking frogs. b. Into that cave continue to crawl strange-looking frogs.


49

c. *Into that cave continue to seem to crawl strange-looking frogs. d. *Into that cave are believed to continue to crawl strange-looking frogs.

7.3.3.12.   In transitive structures, a typical particle can appear either before or after a direct object: (138)

a. b. c. d.

Henry Henry Henry Henry

figured out the answer. figured the answer out. finished off the vodka. finished the vodka off.

However, in presentational there structures the P-subject must always follow a particle. Consider first intransitive cases: (139)

a. b. c. d.

In the Senate, there will come about a reasonable compromise. *In the Senate, there will come a reasonable compromise about. In that house, there grew up a future president. *In that house, there grew a future president up.

Note that the violations here cannot be attributed to some heaviness constraint, as objects identical to those in (139) can precede particles in direct object structures: (140)

Henry picked a future president up (at his house).

Even more strikingly, P-subject particle order is impossible in passives, even when that order is fine in the corresponding actives with the same lexical/semantic items: (141)

a. b. c. d.

To Henry they handed over the key documents. To Henry they handed the key documents over. To Henry there were just handed over the key documents. *To Henry there were just handed the key documents over.

(142)

a. b. c. d.

She cut out several figures from that picture. She cut several figures out from that picture. From that picture there had been cut out several figures. *From that picture there had been cut several figures out.

The generalization that bars P-subject + particle order holds as well for NEX clauses. That is, the patterns of well-formedness/ill-formedness in (139)–(142) are preserved when the expletive there is suppressed from each example that contains it. 7.3.3.13.    (1993) In her exceptionally extensive study of English verb classes, Levin (1993) provides significant information about

50


verb classes that occur with what she calls ‘There Insertion’ and what she calls ‘Locative Inversion’. As far as I can determine, every class Levin lists that permits one of these permits the other and every class she lists that precludes one precludes the other. These are: (143)

a. Page 237 verbs of substance emission (e.g., drip) allow both. b. Pages 245–246 other alternating verbs of change of state (e.g., defrost) preclude both. c. Pages 247–248 verbs of calibratable changes of state (e.g., fluctuate) preclude both. d. Page 248 lodge verbs (e.g., dwell) preclude both. e. Pages 249–250 verbs of existence (e.g., remain) allow both. f. Pages 250–251 verbs of entity-specific modes of being (e.g., froth) allow both. g. Pages 251–252 verbs of being that involve motion (e.g., vibrate) allow both. h. Page 252 verbs of sound existence (e.g., resonate) allow both. i. Pages 253–254 swarm verbs (e.g., swarm) allow both. j. Pages 255–256 verbs of spatial configuration (e.g., perch) allow both. k. Page 256 meander verbs (e.g., stretch) allow both. l. Page 258 appear verbs (e.g., materialize) allow both. m. Page 259 reflexive verbs of appearance (e.g., present [itself]) preclude both. n. Page 260 verbs of disappearance (e.g., vanish) (according to Levin) only weakly allow both. o. Pages 260–261 verbs of occurrence (e.g., happen) allow both. p. Page 262 verbs of assuming a position (e.g., crouch) preclude both. q. Pages 265–266 run verbs (e.g., jog) allow both.

I do not wish to give the impression that Levin’s data reveals a perfect correlation between verbs that permit a there construction and those that permit a corresponding NEX structure. The correlations cited in (143) do not support such a strong conclusion because they are stated in terms of verb classes and her account does not indicate whether both are possible or impossible for each verb of the various classes. In certain cases, she specifically indicates that constructions are possible only for some members of a class. Nonetheless, the data in (143) is still quite strong support for a systematic connection between there and NEX structures of the sort that only (4a) among the analyses here at issue recognizes. This point is supported further by the fact that in certain cases I disagree with Levin’s categorizations. For instance, on page 246 she claims that the change of state verb dry is not possible in either there or NEX forms. I disagree, but the key point is that I find both types possible: (144)

On the line (there) were drying some of the most beautiful shirts I ever saw.

Overall then, I believe Levin’s data provides excellent support for the claim that NEX clauses are a variant of there structures. 7.3.3.14.     Various other difficult to classify environments also reveal, I believe, a fundamental connection be-


51

tween NEX structures and P-there ones. So, for example, contrast the two verbs fall and occur, both of which take human to phrases. The former permits a presentational there structure and, correspondingly, a NEX clause. The latter permits neither: (145)

a. To Jenny (there) fell the task of contacting the parents. b. *To Jenny (there) occurred the idea of contacting the parents.

A similar set of correlated contrasts exists for go versus appeal: (146)

a. To Jenny (there) will go the poetry prize. b. *To Jenny (there) will appeal the handsome waiter.

See also happen and matter, which preclude both presentational there and NEX structures, and flow, which allows both: (147)

a. *To Jenny (there) happened all sorts of terrible things. b.*To Jenny (there) matter many things that seem unimportant. c. To Jenny (there) flowed all of the profits from that venture.

Further, Napoli (1993: 75) gives the following presentational there data that involves intransitive verbs: (148)

a. b. c. d. e. f. g. h. i.

There strode into town the ugliest gunslinger alive. There went up a cry of protest. There appeared a man in the doorway. There lived a king in days gone by that I must tell you about. There stood a little boy in the corner. There suddenly burst in five policemen. *There telephoned a hysterical victim. *There spoke an imposing woman in favor of ozone. *There painted a woman on the bridge over the Seine.

This distribution of grammatical and ungrammatical structures is exactly matched by corresponding NEX cases (I ignore Napoli’s “??” vs. “*” markings in (149), which do not correspond to anything in my judgments): (149)

a. b. c. d. e. f. g. h. i. j.

Into town strode the ugliest gunslinger alive. In the hall went up a cry of protest. In the doorway appeared a man. In days gone by lived a king that I must tell you about. In the corner stood a little boy. At that point suddenly burst in five policemen. *In the corner cried a little boy. *From the scene telephoned a hysterical victim. *At the meeting spoke an imposing woman in favor of ozone. *Throughout the storm painted a woman on the bridge over the Seine.

52


7.3.3.15.   A notable characteristic of both NEX and presentational there clauses is that they are incompatible with at least a wide range of idiomatic objects, even those that are passivizable: (150) (151)

(152)

a. b. a. b. c. d.

At that reunion a great deal of money/attention was paid to the new director. At that reunion (there) was paid to the new director a great deal of money/*attention. During that week (some) strong beverages were kept near the children. During that week (there) were kept near the children (some) strong beverages. During that week very close tabs were kept on the children. *During that week (there) were kept on the children very close tabs.

a. During that exhibition a very strong impression was made on the paper/the audience. b. During that exhibition (there) was made on the paper/*the audience a very strong impression.

7.3.3.16.   A striking and as far as I know never previously observed fact about both presentational there and NEX structures is that they seem broadly incompatible with at least many metaphorical interpretations of verbs. Thus for a verb that has both literal and metaphorical interpretations and permits passives of both, the corresponding presentational there and NEX structures nonetheless may well permit exclusively the literal readings: (153)

a. b. c. d. e. f.

At that orgy many innocent tourists were killed. At that orgy a lot of time was killed in arguments. At that meeting a lot of fine proposals were killed. At that orgy (there) were killed many innocent tourists. *At that orgy (there) was killed a lot of time (in arguments). *At that meeting (there) were killed a lot of fine proposals.

(154)

a. At that meeting several women were screwed by licensed physicians. (i) The physicians had sex with the women. (ii) The physicians did something rotten to the women. b. At that meeting (there) were screwed a large number of women. (i) Unspecified people had sex with the women. (ii) *Unspecified people did something rotten to the women.

(155)

a. On Friday the Yankees were massacred by the Red Sox. (i) The Red Sox murdered the Yankees in a violent way. (ii) The Red Sox beat the Yankees by a wide margin in a game. b. On Friday (there) were really massacred several minor-league teams. (i) Unspecified murdered the members of several minor-league teams. (ii) *Unspecified beat several minor-league teams by wide margins in games.

(156)

a. At that meeting several officials will apparently be roasted. (i) The officials will be cooked. (ii) The officials will be strongly criticized.


53

b. At that meeting (there) will apparently be roasted several notorious officials. (i) The officials will be cooked. (ii) *The officials will be strongly criticized. (157)

a. During that barbecue an enormous number of beers/crazed ideas were swallowed. b. During that barbecue (there) were swallowed an enormous number of beers/*crazed ideas.

It is bizarre that such constraints should have to be stated once in a grammar. But a view that fails to link NEX clauses to presentational there ones would seem committed to stating them twice.27 7.3.3.17.  Bresnan (1994: 103) claimed: (158)

“In summary, we have seen that many locative inversions in English lack a plausible expletive source, that a null subject leads to loss of generalizations over subject extractions, that English lacks other characteristics of null-subject languages, and that a null expletive cannot in any event explain the contrasts between English and Chichewa. For these reasons, I reject the expletive subject hypothesis for English locative inversions.”

But it has been shown here that talk of ‘sources’ is irrelevant, that the subject extraction generalizations cannot be stated over S3s, that the claimed lack of null expletive subject characteristics is only partially true, and that the evidence that links NEX clauses to presentational there clauses is broad, diverse, and sufficiently thoroughgoing as to undermine any account that fails to systematically relate these two types of structures. But that is exactly what an analysis like (4c) does. Only a null expletive treatment of NEX clauses seems to have a chance of capturing their similarities with presentational there structures.28 One is left then with the issues of the relation between English NEX clauses and structures in other NLs and the unresolved issue of the proper way to state the extraction constraints that led Bresnan to the view that X-PPs were subjects (S3s). Before dealing with the latter obviously crucial issue, I turn to the foreign NL question and argue that, with respect to the choice between (4a, b, and c), it cannot cut the way Bresnan has claimed.

8. Foreign tongues One of Bresnan’s (1994) arguments for a subject analysis of X-PPs depends on crosslinguistic considerations—in particular, similarities between English NEX cases, more precisely the locative inversion subpart of this construction, and certain structures from the Bantu language Chichewa, which Bresnan takes to instantiate the same universal locative inversion pattern. Apparently, Chichewa locative inversion structures parallel NEX cases in a number of specific properties and yet it is, according to Bresnan, clear that the corresponding locative forms are subjects. Hence the argument is that only the subject analysis of the X-PPs of English NEX cases captures the generalizations that cover both English and Chichewa.

54


If one accepts the premises about Chichewa locative inversion cases and parallelisms with English NEX cases, this argument might seem to have a good deal of force. Its logic is roughly this. There are two parallel constructions, English NEX cases and Chichewa locative inversion cases in which locative constituents behave in special ways. In English, the status of these constituents, that is, X-PPs, is unclear and controversial, as we have seen. But in Chichewa, their subject status can be taken as established. This supports the subject status of X-PPs, since it permits stating the generalizations about the specific properties as properties of locative inversion subjects. But the ability of this line of argument to show something about English is dependent on many other factors. An objector could well claim that the parallelisms between English NEX clauses and Chichewa locative inversion clauses are at least partially fortuitious or due to factors that can be captured without taking English X-PPs to be subjects. This sort of objection could be justified if one could find other NLs that have constructions that also share properties with English NEX clauses but where it is clear that their analogs of X-PPs are not subjects. I claim that French is such a language, as illustrated initially in (159): (159)

Examples essentially from Gross (1975: 93–94) a. Un certain nombre d’ennuis résulteront de ta décision. “A certain number of problems will result from your decision” b. De ta décision résulteront un cetain nombre d’ennuis. “From your decision will result a certain number of problems” c. Cette question correspond à cette réponse. “This question corresponds to this response” d. A cette question correspond cette réponse. “To this question corresponds this response” e. Des primes s’ajouteront à ce salaire. “Bonuses will be aded to that salary” f. A ce salaire s’ajouteront des primes. “To that salary will be added bonuses” g. Une chemise à fleurs sort de son pantalon. “A flowered shirt is sticking out of his pants” h. De son pantalon sort une chemise à fleurs. “From his pants is sticking out a flowered shirt” i. Des dossiers sont disposés sur la table. “Files are placed on the table” j. Sur la table sont disposés des dossiers. “On the table are placed files” k. Des personnes bien disposées interviendront à ce moment-là. “Well-intentioned people will intervene at that moment” l. A ce moment là interviendront des personnes bien disposées. “At that moment will intervene well-intentioned people”

This construction is also found with passives and interacts with raising in the same way as NEX cases:


(160)

55

a. Des fleurs ont été mises sur la table. “Flowers have been placed on the table” b. Sur la table ont été mises des fleurs. “On the table have been placed flowers” c. Sur la table semble avoir été mises des fleurs. “On the table seem to have been placed flowers”

Surely, in examining cases like these, it is hard to imagine that this French construction is not at least as close to English NEX structures as Chichewa locative inversion cases are. If so, then valid cross-linguistic generalizations should treat French and English structures in parallel ways. But it seems impossible to analyze French cases like (159) and (160) in a fashion parallel to the way Bresnan analyses NEX structures, that is, with the preverbal PP treated as a subject (S3). For French also has an analog of the English Anticomplementizer Constraint; as a body of work makes clear, this is represented by the socalled que/qui alternation; see, for example, Kayne (1976; 1984: 69–71; 94–98); Moreau (1971); and Pesetsky (1981/1982). That is, where the standard French equivalent of the English finite complementizer that is que, the complementizer qui is required in cases of straightforward subject extraction: (161)

a. l’homme que je crois que/*qui Marie déteste “the man that I believe that Marie detests” b. l’homme que je crois *que/qui déteste Marie29 “the man that I believe detests Marie”

If then it were appropriate to analyze the French analog of NEX structures as involving PP subjects, one would expect that these cases would, like their English analogs, determine the Anticomplementizer Constraint, that is, behave like subject extractions. But they do not: (162)

a. la table sur laquelle je crois qu’/*qui ont été mises des fleurs “the table on which I believe have been placed flowers” b. A quelle question croit-il que/*qui correspond cette réponse? “To what question does he think corresponds that answer”

Thus if one adopts the very logic Bresnan appeals to and insists that English must be analyzed parallel to the straightforward analysis of an NL that contains a construction that parallels NEX structures, one is led to reject Bresnan’s subject analysis of X-PPs in favor of an analysis that is also applicable to French. This is an analysis that treats the French cases as involving null expletives and hence only nonsubject extraction in examples like (162). Since it is surely no less obvious that the French/ English parallels are genuine than that the Chichewa/English ones are, at best the French facts undermine any argument for the analysis of NEX clauses based on Chichewa. If, as I believe, it makes much more sense to treat the French and English constructions the same than it does the Chichewa and English ones, the French facts provide a positive argument against Bresnan’s analysis of NEX clauses.

56


And the French situation suggests in another way that a key problem in the treatment of NEX cases is the proper formulation of the English Anticomplementizer Constraint. One now sees that this is not completely parallel to the French Anticomplementizer Constraint in that in English, but not in French, the relevant effects are found in structures like NEX clauses. One key question that needs to be addressed then is what factor differentiates English NEX clauses from their French parallels in such a way that the former yields the Anticomplementizer Constraint, but the latter does not. Nothing in Bresnan’s proposal addresses this issue, which is treated at the end of the following section.

9. A novel account of NEX structures 9.1. Capturing the similarities Previous sections have, I believe, shown that the X-PPs of NEX structures are not S3s and that such clauses have invisible expletive S3s of the there type. In these terms, the failure of X-PPs to manifest most subject properties and the broad sharing of properties between NEX clauses and presentational there clauses that has been supported are entirely expected. Not expected is that NEX clauses obey the three extraction constraints of section 2. It has already been argued that the apparently conflicting implications of the cooccurrence of these two sets of facts are not paradoxical, inter alia because, as independently supported by the A-Parenthetical data, the extraction constraints cannot be uniquely defined over S3 extraction. But that has left open just how they can be defined in such a way as to capture the English-internal generalizations. And the previous section has revealed that it is in addition reasonably requisite that the proper solution for that also provide a means of specifying that the French analog of NEX clauses does not yield in particular (the analog of) the Anticomplementizer Constraint. My suggested solution to these problems, in effect an answer to Bresnan’s (1994) claim that an expletive S3 analysis of NEX clauses fails to account for the extraction data, appeals to the following logic: The (extracted) X-PPs of NEX clauses, unlike all other extracted PPs, are treated by the three extraction constraints of section 2 like extracted S3s despite not being extracted S3s, because X-PPs share certain abstract grammatical properties with S3s, properties that are not shared with other constituents like direct objects and arbitrary other PPs. These properties can, I suggest, be characterized within an appropriate theoretical framework in terms of a view of the abstract syntactic structure that underlies case. It seems fair to say about both traditional notions of case marking and modern applications and extensions of them that at least two different levels of ideas are involved. First, there is a syntactic aspect. At this level, one can for example recognize that there is some close relation between subject and nominative case, between direct object and accusative case, and so forth. On a different level, there is the question of the morphological nature of case marking, whether it is even morphologically present and, if so, where, whether it involves prepositions, postpositions, infixes, inflections, and so on. It should not be terribly controversial to hold that these two aspects are in principle independent. I have nothing to say about the morphological aspect here. My concern is exclusively with the syntactic aspect. Since my notions about this are quite nonstandard, I take the never-pleasing step of denoting the rele-


57

vant concepts by a novel term, Quace. It will ultimately be claimed that each of a range of core grammatical relations, subject, direct object, indirect object, and such, defines a syntactic object called a Quace and, moreover, that these objects can be associated with certain other constituents (e.g., complementizers) via a type of agreement. The idea will be that, for example, final subjects determine for themselves a default type of Quace, call it subject-Quace (hereafter1-Quace and, in general, Quace types will be defined by the R-signs of [168] later), but that final subjects can also have a partially distinct Quace, and that elements that are not final subjects can have a type of 1-Quace as well. With respect to NEX structures, application of the ideas just invoked can be introduced highly informally as follows: Viewed from the vantage point of other phrases, both X-PPs and the invisible expletives of NEX clauses I have argued for are unusual. The former are unusual in two respects: They are obligatorily extracted and, although non-S3s, are subject to the three constraints otherwise largely but not completely characteristic of extracted S3s. The latter are unusual in that they are obligatorily invisible, whereas, of course, other instances of expletive there not only need not be but also cannot be: (163)

a. b. c. d.

There suddenly broke out violent disagreements. *Suddenly broke out violent disagreements. Harriet prevented there from being an argument. *Harriet prevented from being an argument.

Now, it is known that there is a domain of unusual or special behavior for nominal expressions instantiated in particular under notions of noncanonical case marking; see Aikhenvald, Dixon, and Onishi (2001) for a recent survey of this domain. Normally, talk of noncanonical behavior is invoked in cases where a nominal expression has an unexpected or quirky case, for example, when a subject is dative or an object is nominative. But if one hypothesizes that there is a specific type of syntactic structure that underlies the relevant sort of case marking as well as more expected kinds (call the latter straight cases), one can appeal to a more general notion of quirky behavior even in contexts where no morphological case marking is present. This is the idea that underlies my proposal about English NEX constructions. Specifically, if the syntactic structure that underlies case marking is characterized in terms of the notion Quace, to be elaborated, then the idea is that there is, more fundamentally, straight and quirky Quace. The assumption then is that while the expletives I have argued are S3s of these constructions are subjects, they are quirky Quace subjects, that is, subjects with a Quace distinct from 1-Quace, the quirkiness being what underlies their nonstandard property of invisibility. In an NL with case marking, one would in such terms expect that the analogs would have some quirky case, that is, a non-nominative one. Just so, while the X-PPs of NEX constructions are not S3s and thus would not in standard terms have the structure that determines nominative marking, I assume that they, too, have quirky Quace, in particular, quirky 1-Quace. This will mean that they exceptionally possess in part the sort of overlain syntactic structure that determines nominative case on S3s regularly. In these terms, NEX clauses represent two types of jointly cooccurring quirky nominal structures; in addition, it must be assumed that

58


by and large each of them is only possible in the presence of the other. The solution of the paradox developed earlier, with X-PPs sharing some (the extraction) properties with S3s but contrasting with true S3s in a host of other features, then takes the following form. The extraction constraints of relevance, which treat X-PPs and subjects (S3s) alike, will be claimed to be defined not directly over (final) subjects (S3s) but over a slightly larger domain, namely, that of phrases with a restricted type of 1Quace. It can then be claimed that the elements of A-Parentheticals and comparatives, which were shown to also obey some of the extraction constraints that X-PPs and S3s do, also have 1-Quace. More precisely though still obviously totally informally, in English all and only S3s, X-PPs, A-Parenthetical as forms, extracted comparative elements, and fronted adjectival and participial phrases, which also induce the Anticomplementizer Constraint (see [179] later), will have 1-Quace. 9.2. A few generalities about the Metagraph framework To develop the ideas alluded to so far in a more precise (but still highly sketchy and incomplete) way, I will appeal to a development of an arc-based conception of syntactic structure like that of Johnson and Postal (1980) and Postal (1986a, 1990a, 1992, 1996), which, following Postal (1992), I refer to as Metagraph Grammar. The basic Metagraph idea about syntactic structure is that sentences are built of objects called arcs, nodes, and two primitive grammatical relations between arcs called Sponsor and Erase. Each arc represents the existence of a grammatical relation that holds between the object that is the head of the arc and the object that is the tail of the arc. Which relation an arc represents is indicated by its label, called a R(elational)-sign. That is, an arc is a pair that consists of a primitive edge plus an R-sign. Each edge is paired with two nodes, one a head node, one a tail node.30 The following remarks hold for the two primitive relations between arcs. They are each binary and irreflexive. Moreover, some arcs are sponsored (stand as second arguments of the Sponsor relation), while others are not; the latter are called initial arcs. Sponsored arcs have exactly one sponsor. If arc A sponsors arc B and they have the same head node, then B is a successor of A (reciprocally, A a predecessor of B). Any relation between arcs A and B is local if and only if A and B have the same tail node, meaning their head nodes are sisters; such arcs are called neighbors. Otherwise a relation between arcs is foreign. For any relation R between arcs, the logical ancestral (a reflexive relation) of R is denoted by Remote-R. So, for instance, if D is a successor of C, C a successor of B, and B a successor of A, then each of A, B, C, and D is a remote successor of itself, D is in addition a remote successor of A, B, and C, C is in addition a remote successor of A and B, and B is in addition a remote successor of A. Some arcs have erasers (that is, stand as the second arguments of the Erase relation); others do not. No arc has more than one eraser. The Erase relation plays a key role in picking out from the total set of arcs in a sentence V a subset that represents the superficial structure of V. A necessary condition for membership in this subset of arcs is that an arc not be erased. To illustrate a bit, assume that a clause like (164) involves an unaccusative verb, which in Metagraph terms means one whose only initial argument, here realized as Dragons, heads an (initial) 2-arc:


(164)

59

Dragons die.

Since that phrase is a superficial subject, the overall structure of the clause would include at least the elements of (165, fig. 1.1): (165)

A 1

B 2

C P

Dragons die

 1.1

In (165) and later, the Sponsor relation is graphically represented by a wiggly arrow with the sponsored arc at its point, the Erase relation by a double arrow with the erased arc at its point. Here arcs B and C have no sponsors, while B sponsors A. By the definition, the 1-arc A here is a local successor of the initial 2-arc B, also a remote (local) successor of it. In this simplified representation, A and C would qualify as superficial arcs, since both lack erasers, but B would not, as it does have one, namely, A. In (165), A, the successor of B, erases its predecessor. From a purely logical viewpoint, nothing precludes the inverse situation, erasure of a successor by its predecessor. While this pattern was claimed to be impossible in Johnson and Postal (1980), subsequent work, which included Postal (1992, 1996), has assumed that such relations are possible. And that assumption is crucial for specification of the notion of Quace presented here. I thus distinguish at least two subtypes of successor relations: (166)

Definition a. A is a type I successor of B (reciprocally, B a type I predecessor of A) if and only if A is a successor of B and A erases B. b. A is a type II successor of B (reciprocally, B a type II predecessor of A) if and only if A is a successor of B and B erases A.

Given what was said about the relation between the Erase relation and superficial structure, it can be seen that type I predecessors, like B in (165), are never candidates for inclusion in superficial structure. Therefore, type I successor/predecessor pairs can represent inter alia different types of ‘changes of status’; one instance is that of a phrase

60


that, like Dragons in (165), is an object at one level of a fixed constituent but a subject at another ‘later’ one, with only the ‘later’ status having the possibility of determining a superficial status. Another is that where a phrase that is an initial and final subject of one clause ‘raises’ into a higher one, which, in Metagraph terms, indicates that that constituent will head an arc that is a type I foreign successor of the final 1-arc it heads. This means then that the constituent can be a superficial constituent of the higher clause but not of the lower one, although it is a constituent of both. However, for type II predecessors, no such remarks hold. For an arc A to have a type II successor has no direct implication at all for the potential superficial status of A. One might say then, entirely informally, that type II successor structures provide a sort of ‘covert’ status for a phrase, one that not only is not but also cannot be directly represented in its surface status. So, heading a type I successor arc whose R-sign is Q gives a phrase the possibility of being a surface Q. But heading a type II successor arc whose R-sign is Q does not; it merely offers the possibility of sharing some properties with surface Qs. A crucial notion for this discussion is that of final arc. I will, generalizing a bit the statement in Postal (1992), take a final arc to be one that has no eraser arc internal to the constituent defined by its tail node, but see later comments and also note 35. More precisely: (167)

Definition An arc A is final if and only if no Remote Branch of any neighbor of A erases A.

Since A is a branch of B if and only if A’s tail node is identical to B’s head node, a final arc is defined as one that is either not erased at all or erased by some arc that cannot be reached from A by ‘traveling’ down branches that start at A or any neighbor of A. I assume further that there is a class of Nominal-arcs, defined by their R-signs, which includes at least those of (168). Hereafter, arcs that represent the relevant relations are named by their R-signs; so arcs that represent the subject relation are 1-arcs, and so on: (168)

Nominal-arc R-signs and the Relations They Characterize 1 subject 2 direct object 3 indirect object 4 subobject (see Postal, 1990a) 5 semiobject (see Postal, 1990a) 6 quasiobject (see Postal, 1990a) 8 chomeur 9 extraposed clause Oblique1 Oblique2 ... Obliquen

My assumption is that it is (only) (possibly a subset of) Nominal-arcs that define the domain of ‘primitive’ Quace determination. Other types of arcs will, as touched on presently, receive Quace, but only via what I take to be agreement phenomena.


61

9.3. The notion of Quace My assumption is that each final Nominal-arc determines a default Quace structure. Specifically, a final 2 should determine default 2-Quace, a final 3 default 3-Quace, and so on. To define such a default structure, I see at least two possibilities. Under one, call it Proposal A, the default Quace of a final R-arc A could simply be taken to be some equivalent type II local successor of A, where two arcs are equivalent if and only if they have the same R-signs. Under another, Proposal B, the default Quace of a final R-arc A could be induced from A itself. That is, one could introduce a definition in which a final R-arc has by definition R-Quace, with no need to postulate an equivalent type II local successor for A, as in Proposal A. Proposal A might seem clumsy and “redundant,” as it requires every final Nominal-arc to have an equivalent type II local successor. Moreover, it requires a condition like (169): (169)

Condition 1. Equivalent Quace If A is a final Nominal-arc, then there is a B that is an equivalent type II local successor of A.

Proposal B would not require (169). These formal factors indicate that it is worthwhile trying to develop an account of Quace in terms of Proposal B. But attempts along that line have not been very successful and thus here I adopt Proposal A. To understand the issue, it is necessary to recognize that ultimately it seems correct to take the Quace structure of a final arc A to consist of a sequence of arcs, certainly including some type II local successors. This fact makes Proposal A seem more motivated than it would simply in isolation because, in contrast to Proposal B, it permits the sequence of arcs to consist of nothing but type II local successors, the simplest situation. One can take advantage of the ancestrals of basic relations such as RemoteSuccessor, to define Quace structure as follows internal to Proposal A: (170)

Definition: Quace Marker A sequence of arcs {Q1. . . Qn} is the Quace Marker of an arc A if and only if a. A is not a type II local successor, and b. A is a type II local predecessor of Q1, and c. Qn is not a type II local predecessor, and d. for all i, 1 ≤ i ≤ n, Qi+1 is a Nominal-arc type II remote local successor of Qi.

The key points about a definition like (170) are, first, that it does not determine that an arbitrary arc has a Quace Marker. However, second, combined with Condition 1, it does determine that every final Nominal-arc has one. In such cases, the Quace Marker of a final arc A consists of the maximal sequence of Nominal-arc type II remote local successors of A. In no case can a Quace Marker be null. Third, while Definition (170) determines that the Quace Marker of a final Nominal-arc A could consist exclusively of B, the single equivalent type II local successor of A whose existence Condition 1 guarantees, it also allows for richer possibilities in which B itself has a type II local successor.

62


The former situation in which there is a Quace Marker with one member will be taken to define what was called the default situation, that is, straight Quace. The latter possibility can be taken to define the notion quirky Quace. Put differently, one can specify that a particular type of final constituent in a particular context has quirky Quace by specifying that the final arc A that defines such a final constituent has not only the equivalent type II local successor determined by Condition 1 but also at least one nonequivalent type II remote local successor whose R-sign is distinct from A’s. Under the assumption that the notions relevant to Quace are limited to the domain of nominals, (170) limits all members of Quace Markers to Nominal-arcs. The exclusion of A itself from this specification is needed, as will be seen, to allow, for example, the arcs headed by complementizers, which are not Nominal-arcs, to have Quace Markers, a possibility instantiated, I suggest, only through agreement in Quace. Possibly, more generally, Quace Markers for non-Nominal-arcs of any type arise only through Quace agreement. Many questions evidently arise about Quace Markers, but the search for their answers is in general beyond the scope of these remarks. Certain of these questions concern more generally type II local successors. One is whether a single arc can have more than one. I suggest not, a state of affairs imposed by Condition 2. (171)

Condition 2. Type II Local Successor Uniqueness If A and B are type II local successors of C, then A = B.

No analog of Condition 2 is necessary to exclude the case where one arc has two type I local successors, excluded by the uniqueness of the Erase relation. That is, an arc that has two type I successors, local or not, violates the Unique Eraser Law of Johnson and Postal (1980), which limits each arc to a maximum of one eraser. While Condition 1 specifies that each final Nominal-arc has a relationally equivalent type II local successor, it does not itself specify that no final Nominal-arc has any relationally nonequivalent type II local successor. That is, it requires situation (172a, fig. 1.2a) but would alone be satisfied by (172b, fig. 1.2b): (172)

a.

A B

A is a final Nominal-arc R

R

 1.2a


63

b. C B

A

R

Q

R A is a final Nominal-arc

 1.2b

But (172b) could exist only if a type II predecessor could have distinct type II local successors, which, so far purely on grounds of conceptual neatness and lack of known instantiations, is barred by Condition 2. Another question is whether type II local successors can themselves have type II local successors. The development here ends up requiring this. But I propose minimal limitations on this situation, in particular, eliminating the case where a type II local successor has an equivalent type II local successor. This can be done more generally by requiring cases where a local successor and its predecessor have the same R-sign to be such that the predecessor is not a type II local successor and its successor is. (173)

Condition 3. Local Successor Equivalence If A is an equivalent local successor of B, then A is a type II local successor and B is not.

This excludes such situations as (174a, b, figs. 1.3a, b): (174)

a. An equivalent type I local predecessor/successor pair:

A

R

R B

 1.3a

64


b. An equivalent type II local predecessor/successor pair where the predecessor is also a type II local successor. A

B

C

R Q

Q

 1.3b

Condition 3 not only bars a type II local successor from having an equivalent type II local successor; it also bars the apparently never motivated and redundant situation, which unhappily did not follow from anything independent in the system of Johnson and Postal (1980), where an arc has an equivalent type I local successor. This state of affairs would be consistent with arbitarily long sequences of useless local successors like B in (174a), none of which can influence the surface status of a constituent. Other questions that arise from the notion defined in (170) are specific to Quace Markers. For instance, one wants to know about bounds on their size, on the possibility of the same R-sign being repeated in the sequences (though given Condition 3, necessarily at best only in nonadjacent Quace Marker positions), and so on. Returning to descriptive issues, the intention is that in these terms, a quirky case, say dative, final subject would instantiate substructure (175, fig. 1.4), under the plausible view that dative case is determined by 3-arc Quace: (175) A

B

C

1 1

3

D Case

 1.4.


65

In a structure like (175), A’s Quace Marker = [B, C] contains two distinct arcs. Let us say in such circumstances to simplify discussion that for a final arc A whose Quace Marker contains arcs B, C, and so on, that B and C are Quaces of A and also that if B is an R-arc, that A has R-Quace. In a multiple arc Quace Marker, the different arcs do not have symmetrical status and it is important to distinguish them, which one can do terminologically as follows: (176)

Definition A member, A, of a Quace Marker M is finished if and only if A has no (type II) local successor (that is, is the final arc of M).

So in (175) C is finished, but B is not. Structure (175) sketches a possible view of the relation between case and Quace. Although an investigation of the morphological structure of case is (far) beyond this discussion, the analysis in (175) has case represented in part by an arc (here D) with R-sign Case, where D is sponsored by an arc of which it is a branch. Moreover, the sponsor is a Quace-arc and, further, a finished Quace-arc. One can speculate that these conditions are lawful. It would, I think, be bizarre in a structure otherwise like (175) to have D sponsored by, for example, B. For that would mean that the case is unrelated directly to the quirky Quace-arc. Under the narrower view, it might then be possible to simply define the core cases, at least nominative, accusative, and dative, as those sponsored by finished Quace-arcs with, respectively, the R-signs 1, 2, and 3. With the apparatus just (very partially) specified in hand, one can approach the descriptive conclusions about NEX clauses argued for in earlier sections. What I want to say is that English allows a variety of final Nominal-arcs, which include 5-arcs, 6-arcs, 8-arcs, Oblique1-arcs, . . . Obliquen-arcs, to have quirky 1-Quace, that is, to have Quace Markers with a finished Quace 1-arc. But this is subject to several conditions. The first, is that the relevant phrases are ultimately extracted, which, in this framework, means those final arcs are remote foreign predecessors of certain arcs (called Overlay-arcs in Johnson and Postal [1980]), which include Wh Q-arcs, WhRel-arcs, Topic-arcs, and such. The second is that the final Nominal-arc with quirky 1-Quace is a neighbor of a final 1-arc that itself has some kind of quirky Quace. I will specify with some arbitrariness that this is 4-Quace.31 Third, it is assumed that a final 1-arc with 4-Quace is only possible in this construction if it is a neighbor of a final Nominal-arc with quirky 1-Quace. Fourth, I assume that the final 1-arc with 4-Quace, which must be a sister of the arc with quirky 1-Quace, is headed by an expletive and, moreover, exceptionally requires its head node to have a null phonological realization.32 This is still not sufficient to characterize the NEX construction. It must, in addition, be specified that the relevant expletive is there and not one of the expletive it found in English. A full specification of the latter condition would require a full understanding of the difference between expletive there and the various expletive it. I do not possess such. However, it may be possible to specify that a there type expletive arc is one that is a remote successor of a 2-arc sponsored by a neighboring 2-arc whose head represents a DP that is not a complement clause. Therefore, the NEX clause defining structure will include at least the elements of (177) fig. 1.5:

66


(177) A R

R B

F

E

D

4

1

1 1 C

X-PP

there A is a final Nominal-arc

 1.5.

In this structure, I artificially ignore, as I will in what follows, the fact that X-PPs are PPs. This matter raises a variety of issues about the description of PPs that, though important and relevant, are not central to this discussion, so I will not seek to deal with them here. In (177), arc A is the final R-arc, which determines that X-PPs are not S3s. B is the type II local successor required by Condition 1, which represents default Quace, and C is the finished Quace-arc that specifies that A has quirky 1-Quace. E is the default Quace-arc required by Condition 1, while F is the arc that represents that D has finished quirky 4-Quace. I will assume that the conditions just discussed on NEX clauses are a consequence of several distinct principles of English grammar. First, there is a specifically NEX clause characterizing condition that I take to specify three necessary conditions for a final Nominal-arc, like A in (177), which is not a 1-arc, to have finished 1-Quace: (178)

Condition 4. NEX Clause Constraints If A is a final Nominal-arc but non-1-arc, then A’s Quace Marker’s final element is a (finished) 1-Quace arc only if (i) there is a distinct neighboring final expletive 1-arc, B with finished 4-Quace; and (ii) A’s R-sign is a member of {5, 6, 8, Oblique1, . . . Obliquen}; and (iii) A is a type I remote foreign predecessor of an Overlay-arc.

Notable about (178) is that it does not restrict the existence of final expletive 1-arcs with finished 4-Quace to NEX clauses and is consistent with their occurring in any sort of clause whatever. This is obviously inadequate, since I assume that it is precisely finished 4-Quace on a 1-arc that leads to an expletive subject not being pronounced. I will propose, however, that the conditions needed here are relevant more generally than to NEX clauses and are needed at least for those as clauses and comparative clauses shown in section 5 to also involve null expletives.


67

Moreover, I assume that the same conditions hold for fronted participial constructions like (179), which induce the Anticomplementizer Constraint just as NEX clauses and the others do but which space precludes discussing in detail:33 (179)

Lying on the table (he is sure) (*that) were two sick gerbils.

I take this to mean that the participial constituent heads a final P-arc, which has quirky 1-Quace. I thus propose: (180)

Condition 5. Quirky Expletive Subject Constraints If A is final expletive 1-arc, then A has finished 4-Quace only if there is a neighboring final arc with finished quirky 1-Quace.

This constraint simply requires that finished 4-quace on an expletive arc only occur in a clause that contains some final arc with quirky 1-Quace. However, (180) fails to embody the informal assumption that all those expletives that head final 1-arcs with quirky 4-quace are not pronounced. So in particular Condition (180) does not specify that the expletive in NEX clauses is null. Typically, in the Metagraph framework, a null realization of some constituent K is represented by the erasure of (all) the arcs headed by K. That view always raises the question of the identity of the relevant erasers, which, in this case, is not obvious. Certain conclusions are possible, though. Since internal to present assumptions, the quirky 4-Quace constituent, that is, the invisible expletive, needs to be associated with a final arc, the relevant eraser of at least one of the arcs headed by the expletive would not necessarily be internal to the minimal clause that contains that expletive. For given the definition of ‘final arc’ in (167), such erasure would mean that the arc was not final. That conclusion is, moreover, seemingly strengthened since, as seen earlier, the expletive in question can arguably raise, meaning that the 1-arc it heads can have a type I foreign successor. This needs to be assumed for raising cases like (181), argued in section 1 not to show that the X-PPs were subjects: (181)

a. To Jerome seem/*seems to have been sent several threats. b. To Jerome *seem/seems to have been sent a threat.

Moreover, the expletives posited for as cases, comparative cases, and predicate ones like (179) arguably also can raise: (182)

a. b. c. d.

Jerome is mad, as appears to have been obvious to everyone. Jerome was angrier than appears to be consistent with his mental health. Lying on the table seem/*seems to have been two sick gerbils. Lying on the table *seem/seems to have been a sick gerbil.

Since X-PPs have been shown not to be subjects, to embed cases like (181) and (182) in the general raising pattern, which involves the raising of final complement subjects, one must assume the invisible expletive has raised. By taking there type expletives to agree in number with the DPs they share a relation with, cases

68


like (182c and d) are regular under an analysis that posits raising of the expletive. That would mean then that the final 1-arc headed by the expletives in the (have) (been) sent clause of (181) has no eraser in that clause. Rather, it would seem that the ultimate expletive arc that is erased would have to be at least as high as the seem/seems clause. The only thing in that clause that is really particular to the NEX clause structure is the X-PP that has extracted through the seem clause to its ultimate overlay position. This might suggest that the eraser of the relevant expletive 1-arc is one of the 30-arcs. But I reject that idea. My assumption is that the invisibility of the there in NEX clauses as well as other English cases of null expletive subjects is, rather, to be assimilated to the general invisibility of final 1 expletives in NLs like Italian and Spanish, which in general occur with nothing like an X-PP. Arguably then, the eraser for all such expletive arcs has to be the P-arc of the highest clause in which they occur. If so, then it is on various grounds necessary to revise the definition of ‘final arc’ so that the sort of erasure just posited does not preclude an arc being final. One way to do this would be to add a specification that the ‘Remote Branch’ referenced in (167) is a Nominal-arc. At any rate, the idea is then: (183)

Condition 6. Invisibility of Expletives That Head Quirky-Quace Marked Final 1-arcs If A is a final expletive 1-arc with finished 4-Quace whose neighboring P-arc is B, then if A is not erased by an arc distinct from B, A is erased by B.34

The motivation for the second ‘if’ clause is precisely the raising instances discussed earlier where final expletive 1-arcs of the sort at issue have type I foreign successors when they are in the infinitival complements of seem clauses. 9.4. Stating the extraction constraints 9.4.1. Remark The claim of this section is that the three extraction constraints of section 2 treat the X-PPs of NEX clauses in the same way that they treat S3s because of the fact that the former, exceptionally, have quirky 1-Quace, while S3s, which represent the heads of final 1-arcs, regularly have (straight) 1-Quace. Logic then requires that the three constraints each be stated in a way that takes advantage of the sharing of 1-Quace, which means that none of the three constraints can be properly stated via exclusive reference to final 1-arcs. The following three subsections sketch analyses of the three constraints in Quace terms. 9.4.2. The Anticomplementizer Constraint The Anticomplementizer Constraint in its English variant says in effect that the complementizer of a nonrelative type finite clause must be null under a specific condition Q. The constraint will then have two sorts of effects, given a background claim that of the various English finite complementizers only that has a null variant. The others, as if/how/if/like/when/whether, do not permit this. Therefore, in any overall context where condition Q holds, the Anticomplementizer Constraint will force the


69

null variant of the that complementizer and will render ungrammatical a clause with one of the other finite complementizers. The pattern will be like that of (184a–f): (184)

a. b. c. d. e. f. g.

No gorilla did he believe *that/Ø would solve the problem. No gorilla did it look *as if/*Ø would solve the problem. No gorilla did he see *how/*Ø solved the problem. No gorilla did he like it *if/*Ø solved the problem. No gorilla did he like it *when/*Ø solved the problem. No gorilla did he know *whether/*Ø had solved the problem. Culicover, (1993a: 104) [On which table]1 were you wondering whether under certain circumstances t1 might have been put the books that you had bought h. That is the tiger which1 they saw how at a key moment t1 had helped his trainer.

However, when, as in the Adverb Effect environment illustrated in Culicover’s (1993a) example (16d) repeated as (184g), or in (184h), subject extraction does not require a null complementizer, compatibility with a complementizer other than that is possible. Leaving aside how the issue of null realization for different complementizers is to be described, the core of the formulation of the Anticomplementizer Constraint then reduces to specification of condition Q. My suggestion is that the basic constraint is that a complementizer, Cn, is in the relevant environment Q if and only if there is some final Nominal-arc A of the highest clause of which Cn is the complementizer, which has two properties: (i) A has finished 1-Quace and (ii) the head constituent represented by A is extracted past Cn. Moreover, I will take the latter to mean that A is the foreign predecessor of a specific type of arc that defines extractions; assume these to be defined, as in Postal (2001a), by the R-sign 30. The idea then is roughly that any ordinary (see the discussion of 40-arcs later for a bit of explication of this hedge) extraction involves a minimum of three types of arcs. First, there is an arc that is not a 30-arc and has a 30-arc type I foreign successor. Call this a Start arc. In cases of nominal extraction, the Start arc will typically be a final Nominal-arc.35 Second, there is a (non-null, possibly unitary) sequence of 30-arcs, the first a type I foreign successor of the Start arc and each subsequent one, if any, a type I foreign successor of the previous 30-arc. Assume that any type I foreign successor of a 30-arc is itself a 30-arc; call this the Uniformity Condition. Finally, there is some arc, a member of the class called Overlay-arcs in Johnson and Postal (1980), that is a type I local successor of the last 30-arc foreign successor. I will also assume, quite crucially, that extraction always involves successive immediate constituencies; more precisely, an extracted element must be a constituent of every constituent between its original and final positions. So: (185)

Condition 7. Extraction Limitation If B is a foreign successor of A, then there exists an arc C such that B is a neighbor of C and A is a branch of C.

That is, extractions must be like (186a, fig. 1.6a) and never like (186b, fig. 1.6b):

70


(186)

a. Licit Extraction

B C 30 A

 1.6a b. Illicit Extraction

B D

C 30

A

 1.6b

In (186b) the type I foreign successor of A, B, defining the extraction is not, as required, a neighbor of an arc A is a branch of, since A is a branch of C but only a remote branch of D. The conditions that determine the Anticomplementizer Constraint could be made more compact and cross-linguistically expected if it were possible to state a (hopefully universal) condition that links a complementizer to extractions from the clause it marks. This would be possible if there is in effect a kind of agreement between complementizers and (certain) phrases extracted from their clauses. Such a general agreement pairing has been previously recognized in, for example, Irish and Chamorro; see Chung (1998) and Chung and McCloskey (1987). Moreover, it is also known


71

that there is agreement between complementizers and elements of the clause they mark (e.g., subjects) independently of extraction; see Watanabe (2000) for examples, discussion, and references. The possibility for complementizers to agree thus seems clear. I assume here, moreover, that English complementizers manifest at least agreement in Quace and, further, that this agreement is mediated through the following (hypothesized) informally specified state of affairs: (187)

30-arcs agree in Quace with their predecessors (perhaps better and more generally: If A is a remote successor of a 30-arc and A is a successor of B, A agrees in Quace with B).

The generalization in parenthesis here extends the claim of Quace agreement to successors of A, like, for example, A6 in (191b). Thus the more general version says that a 30-arc agrees in Quace with its Start-arc predecessor, that every remote successor of that 30-arc agrees with its predecessor, and that the ultimate local successor of the last 30-arc in the sequence agrees with the latter in Quace. Thus Quace would be in effect passed up from the Start arc to each intervening 30-arc and finally to the ultimate remote successor of the Start arc, an Overlay-arc. To make sense of claim (187), though, it is requisite to give a precise interpretation of the highlighted word sequence. To do this, one can first define an auxiliary notion for a fixed Quace Marker, as in (188): (188)

Definition If M = {Q1, . . . , Qn} is a Quace Marker, then a sequence of R-signs {R1, . . . , Rn} is the Relational Marker of M if and only if for all i, 1 ≤ i ≤ n Qi is an Ri-arc.

That is, a Relational Marker of a fixed Quace Marker QM is simply the total R-sign sequence defined by the successive R-signs of the successive members of QM. Assuming the notion Relational Marker, one can take a specification of the form (189a) to mean (189b): (189)

a. Arc A agrees in Quace with arc B. b. If B’s Quace Marker = {Q1, . . . Qn}, then the Relational Marker of A’s Quace Marker = {R1, . . . Rn, . . .}.

That is, agreement in Quace between two arcs means that the agreeing arc must have a Quace Marker whose initial sequence of arcs matches in R-sign the full set of arcs of the Quace Marker of the determining arc. At this point then, (187) requires that a 30-arc foreign successor, A, of a Start 1-arc with only straight Quace itself have a type II 1-arc local successor and if A has a (type I foreign) successor, that 30-arc also will have to have a type II 1-arc local successor, and so on. That is, in effect, the Relational Marker of a Nominal-arc Start arc is passed on to each member of any sequence of its 30-arc successors. Then, assuming that complementizers head arcs with the R-sign CT, it is possible to say something like (190):

72


(190)

Condition 8. Complementizer Agreement in Quace A CT-arc A agrees in Quace with an arc B if and only if B is a 30-arc neighbor of A.

Given (190), a simple subject extraction case of the sort relevant to the Anticomplementizer Constraint like (191a) will involve a structure that includes a subpart like (191b, fig. 1.7). To simplify diagrams for comprehensibility, however, here and later I suppress drawing separate type II local successor arcs for 30-arcs and sometimes for CT-arcs as well. Instead, I draw such arcs with complex R-sign combinations of the form 30/1/. . . , CT/1/2, and so on. This should be taken to mean that the first arc has a type II 1-arc local successor and if there is a third symbol, say 2, that 1-arc has a type II 2-arc local successor, and so forth: (191)

a. I know who they claimed (*that) supports Saddam. b.

A6

Wh-Q 1

A5

1

A4

2 P

30 1

30 1

they B2 B1 1

claimed

CT 30 1 A3

A1 1

P

2

A2 1 Saddam supports

that

who

 1.7


73

A nonsubject extraction, say one like (192a), whose Start is a 2-arc, would be a model of (192b, fig. 1.8): (192)

a. I know who they claimed (that) Saddam supports. b.

1

2 P

Wh-Q 2 A6

they 30 2

A5

30 2

B2 2

A4

B1

claimed

CT 30 2 A3

A1

P

1

2 2 A2

Saddam supports

that

who

 1.8

Notably, in (192) the CT-arc ends up via Quace Agreement with finished 2-Quace, not finished 1-Quace, as in (191). At this point, the tools are largely available to indicate initially that both subject and X-PP extraction out of a complement clause must, ignoring the Adverb Effect

74


and another restriction discussed later, yield a null complementizer. For at this point the relevant CT-arcs whose heads must be unpronounced have Quace Markers with finished 1-Quace.36 The descriptive task can then be accomplished apparently by having the relevant 1-Quace arc erase the CT-arc. (193)

Condition 9 Nonrelative Clause CT-arc Erasure A CT-arc of a nonrelative clause with finished 1-Quace is erased by the arc it agrees with in Quace.37

Given (193), the head of CT-arc B1 in (191b) will have to be erased by the 30-arc A3, although this erasure is not indicated there. However, despite what has been said, in at least two subcontexts the description constructed so far fails to work correctly. First, for subjects extracted from a clause deeper than that the CT-arc is directly associated with, there are contrasts like (194a, b): (194)

a. the terrorist who I believe (*that1) threatened Ernie b. the terrorist who I believe (that2) Ernie claimed (*that1) threatened you.

In both (194a and b) the CT-arcs headed by complementizer that1 should seemingly have 1-Quace as a consequence of agreement with the 30-arc, call it A, headed by the relative pronoun who, with that 1-Quace being determined by the fact that the 30-arc is a remote successor of the final 1-arc of the complement of claimed. But the CT-arc whose head corresponds to that2 apparently must not have 1-Quace, although it is a neighbor of a 30-arc that is a remote successor of A and thus, by previous specifications, that is, principles (187) and (190), should agree with it in Quace. In other terms, despite the fact that subject extraction is present, the complementizer that2 in (194b) does not manifest the Anticomplementizer Constraint, and this fact and its contrast with the case for that1 is so far unaccounted for. I suggest one can handle this contrast without altering anything said so far by positing an additional principle of restricted Quace determination by main predicates from whose complements things are extracted. This takes account of the fact that under previous assumptions the 30-arc that is a neighbor of the CT-arc, A, associated with that2, unlike that CT-arc associated with that1, is a (remote) successor of a 30-arc, B, which is a neighbor of a P-arc (one that corresponds to the verb claimed in (194b). I suggest then that the 30-arc B is forced by a further principle to have an additional Quace-arc, which I will hypothesize is a 2-Quace: (195)

Extraction Quace Assignment (first version) If A is a 30-arc whose Start is a final arc with finished 1-Quace and A is a neighbor of a P-arc, then A has finished 2-Quace.

This principle can in general be satisfied by a 30-arc only if the latter’s Quace Marker consists of the sequence {. . . , A, B}, where A is a type II 1-arc local successor and B is a type II 2-arc local successor of A. This is a reason why, as noted earlier, one must allow some type II local successors to themselves have type II local successors.


75

Given this new principle, the 30-arc remote successor of the 1-arc that might have been expected to determine (via the agreement specified in [190]) 1-Quace on the highest CT-arc in (194b) actually determines finished 2-Quace on that arc. Such a CT-arc is then properly not required by (193) to be erased. That is, the relevant parts of (194b) would include the elements of (196, fig. 1.9): (196) 1

P

2 C CT CL

2

30 1 2

I

believes

1

that2

1

P

2

D A6

30 1 2 A5

30 1

Ernie claimed

2

A4 CT 1 B

A3

CL

30 1 that

1 A2

who

A1 1 P

threatened

2

you

 1.9.

Here the CT-arc B must have finished 1-Quace, via agreement with its neighboring 30-arc, A3, which is not a neighbor of any P-arc and thus has no source of 2-Quace. But the 30-arc A4 is a neighbor of the P-arc D and thus must have finished 2-Quace.

76


Hence, by further agreement, A5 has finished 2-Quace, and thus the CT-arc C has it by agreement with A5. In these terms, it is not an accident that the higher instances of that do not yield the Anticomplementizer Constraint, for their CT-arcs are neighbors of 30-arcs that are remote successors of a 30-arc that is a neighbor of some Parc, which requires finished 2-Quace via principle (195). A virtue of the approach to the (194a, b) contrast just sketched is that it likely can be extended to the second so far untreated issue, which is the Adverb Effect. Recall that in the presence of many fronted adverbs, as first noted by Bresnan and stressed in later work by Culicover, the Anticomplementizer Constraint is not found: (197)

a. the nurse who he believes (*that) would watch her father b. the nurse who he believes that under those conditions would watch her father c. *the nurse who he believes Ø under those conditions would watch her father.

Here, though, there are actually two phenomena. First, unlike the ordinary subject case of (197a), the that of the grammatical (197b) is not required to be null. Second, unlike other (e.g., object extraction) cases, the that is not permitted to be null, so that (197c) is ill formed. Why should these properties hold? A partial answer, I believe, lies in the following circumstance, which is partially parallel to that which holds in cases like 194b). In the latter, as represented in (196), the complementizer that fails to behave as if it had finished 1-Quace, that is, that2, heads CT-arc B. As already discussed, the head of B is separated from the position of the head of the Start arc of the 30-arc with which B agrees in Quace by an intermediate verb, which heads some P-arc, there D. I have claimed that that P-arc determines that a neighboring 30-arc with 1-Quace has finished 2-Quace; so A4 is required to have finished 2-Quace. Notably then, in (197b, c) the fronted adverb that nullifies the Anticomplementizer Constraint is intermediate in position between the position of the head of the Start 1-arc and that of the complementizer, which fails to be null. This means that under past assumptions some 30-arc that accounts for the extraction of the 1 will be a neighbor of the arc headed by the adverb. Assuming that the R-sign of this adverbial arc is of a systematically specifiable type, say Z, then one can simply extend (195) to say: (198)

Condition 10 Extraction Quace Assignment (Final Version) If A is a 30-arc whose Start is a final arc with finished 1-Quace and A is a neighbor of a P-arc or a Z-arc, then A has finished 2-Quace.

Given (198), it is now predicted that (197b) is possible, since the complementizer heads a CT-arc whose finished Quace will be 2-Quace, which does not invoke principle (193). The proposal is thus that the fronted adverbs that nullify the Anticomplementizer Constraint also determine non-1-Quace as the finished Quace on a CT-arc just as higher predicates do. The relevant structure of (197b) would be (199, fig. 1.10):


77

(199)

1 P

2

A5

he believes

30 1 2

C CL CT A4 1 2 30 1 2 B

Z A3

that

30 1 2 A1

under those conditions

1

A2 1

who

P

2

would watch

her father

 1.10.

In (199), the key arc is A3, which, as a neighbor of the Z-arc B, is forced by (198) to have an extended Quace Marker with finished 2-Quace. That means that A4, the successor of A3, is forced to have finished 2-Quace, which determines that the CT-arc C has it, guaranteeing that it need not be null. Further support for the view embodied in (198) arises when attention is turned to the Parallelism Constraint and the Complementizer Effect.38 Before turning to the other constraints that extracted X-PPs share with extracted S3s, one should briefly note parallelisms between this description of the Anticomplementizer Constraint and Kayne’s (1984: 6–7) account of what I claim is a related phenomenon. Kayne dealt with the marginal and perhaps largely archaic option of marking a subject interroga-

78


tive or relative pronoun as whom, noting cases like (200a): (200)

a. b. c. d.

the man who(m) I believe has left the man who(*m) left Who(m) do you believe left first? Who(*m) left first?

A key fact is that (200a and b) clearly contrast, as do (200c and d). In Kayne’s transformational terms, cases like (200a and c) were the result of a second cyclic case assignment into Comp. In this framework, the extra objective case marking is arguably possible in just the context seen earlier to determine finished 2-Quace on a 30-arc, that is, in a context where a WH form achieves its final position only by raising past the locus of a predicate (here, believe).39 If the parallelism is genuine, one would also expect that the contexts of the Adverb Effect, unified with the higher predicate context in (198), would permit the case marking seen in (200a, c). I believe this is at least weakly the case: (201)

a. a woman who(*m) might be hired b. a woman who(??)m under the sort of conditions you have just referred to might be hired c. a woman who(??m) at the time the police believe the crime was committed was thought to be in Istanbul d. I don’t know who(*m) might be hired. e. I don’t know who(??m) under the sort of difficult conditions you have just referred to might manage to get hired. f. a politician who(*m) looks ill whenever I see him g. a politician who(?m) whenever I see him looks ill. h. someone who(*m) is now the ruler of the country i. someone who(??m) for all intents and purposes is now the ruler of the country j. someone who(*m) has supported terrorism for years k. someone who(??m), as the National Review has recently documented, has supported terrorism for years

While (201b, c, e, g, i, and k) are hardly lovely, there is, I find, a sharp contrast with the hopeless long versions of (201a, d, f, h, and j). Arguably then, the special objective case marking of an extracted S3 found in (200a) should be limited to the head of an Overlay-arc, which, while a remote successor of a Start arc with finished 1-Quace, is also a remote successor of an intermediate 30-arc, which, via principle (198), has finished 2-Quace. Properly then, this possibility cannot be available for the Overlay-arc headed by whom in (200b), which has no way of receiving finished 1-Quace via principle (198). 9.4.3. The Parallelism Constraint Turn then to the Parallelism Constraint. In its primitive form this prevents a main clause final subject from being combined under coordination with a main clause nonsubject, yielding, for example:


(202)

79

a. *a woman who sang well and Greg praised b. *a woman who Greg praised and sang well

However, the constraint does not bar such combinations if the final subject at issue is an embedded one: (203)

a. a woman who Greg proved could sing well and Tony wanted to meet b. a woman who Tony wanted to meet and Greg proved could sing well

Bresnan’s key observation then was that X-PPs behave in the same way, which she took to support the claim that they are subjects (S3s in our terms here). (204)

a. *In that cage was lying a gorilla and they then placed a monkey. b. In that cage they believed was lying a gorilla and they then placed a monkey.

It has been seen that Bresnan’s conclusion is not viable. Moreover, given the Quace assignment principles of this section, no assumption that X-PPs are final 1s is motivated by facts like (204b). All that is required, arguably, to determine that they behave like 1s under the constraint is that the latter be stated in terms of finished 1-Quace. While the facts of coordination are too complex to permit here a serious account of the Parallelism Constraint, it seems that the following informal statement is defensible: (205)

Coordinate combinations of the sort that underlie Bresnan’s Parallelism Constraint preclude combinations of two constituents, one of which heads an Overlay-arc with finished 1-Quace and the other of which does not.40

Moreover, formulation (205) explains the difference in (204). That is, in (204a) the underlying PP in the first conjunct, which is a NEX clause, will have finished 1-Quace while that in the second conjunct will not. Combination is thus banned. But in (204b) the PP in the first conjunct has no source of finished 1-Quace and neither does that in the second; violation of (205) is not possible. If that is the proper view, then given constraint (198), one would expect that, like the Anticomplementizer Constraint, the Parallelism Constraint should also be nullified in environments where the Adverb Effect holds. This seems correct: (206)

a. *He believes that in that cage they kept a monkey and had already been placed a gorilla. b. ?He believes that in that cage they kept a monkey and at some earlier point in time had already been placed a gorilla.

While (206b) is not lovely, it shows considerable improvement over (206a), which would lack any Z-arc to assign finished 2-Quace to the X-PP that is combined with an ordinary PP from the first complement conjunct. See also the parallel: (207)

a. *Stella claimed that during that era the Arabs had formulated many axioms and had been proved many theorems..

80


b. ?Stella claimed that during that era the Arabs had formulated many axioms and despite considerable controversy had been proved many theorems.

If something like (205) is correct, it predicts facts like (208) under my assumption, necessary given the inducement of the Anticomplementizer Constraint in cases like (179), that fronted participial phrases in inversion cases represent arcs with finished 1-Quace: (208)

a. b. c. d.

Arrested by the police may have been Mickey Cardozo. Arrested by the police Sally Reynolds is said to have been. Arrested by the police is said to have been Sally Reynolds. Arrested by the police Mickey Cardozo may have been and Sally Reynolds is said to have been. e. Arrested by the police may have been Mickey Cardozo and is said to have been Sally Reynolds. f. *Arrested by the police may have been Mickey Cardozo and Sally Reynolds is said to have been. g. *Arrested by the police Sally Reynolds is said to have been and may have been Mickey Cardozo.

In (208d), the participial phrase does not correspond to the predicate of an inversion clause and thus would not head an arc with 1-Quace. In (208e), the participial phrase corresponds to the predicate of inversion clauses in both conjuncts; thus a 1-Quace phrase would be combined with a 1-Quace phrase. But in the ungrammatical case, (208f) the predicate of the first clause would have 1-Quace while that of the second would not, and in the ungrammatical (208g) the predicate of the first clause would not have 1-Quace while that of the second would. Note that in these terms, principle (195) would be irrelevant, since it only determines assignment of 2-Quace to arcs that have 1-Quace, which, given condition (180), is just not the case for the participial phrase arc of the second conjunct of (208f) or the first conjunct of (208g). 9.4.4. The Complementizer Effect Finally, I turn to the Complementizer Effect, which holds only of relative type clauses and which is something of the opposite of the Anticomplementizer Constraint; that is, it requires that a complementizer be non-null: (209)

the vampire that/*Ø attacked Tony

It was observed in section 1 that this also holds for X-PPs: (210)

It was to Larry that/*Ø were sent threatening postcards.

Serious discussion of relative clauses being impossible here, suffice it to say that the fact that this constraint holds for both S3 and X-PP extractions can also be captured by stating the constraint in terms of 1-Quace. Assuming that the relative clause ini-


81

tial that of cases like (210) is a complementizer, agreement principle (190) already determines that CT-arcs like that headed by that in (210) have finished 1-Quace. So an informal version of the constraint needed for relative clauses might say: (211)

If A is a relative clause initial CT-arc and has finished 1-Quace, A is not erased.

If current assumptions are roughly correct, the embedded environments and the Adverb Effect, which both lead to assignment of finished 2-Quace (via condition [198]), should nullify the consequences of (211). This is correct (for the higher complementizer) in the case of the embedded environment (see [212]), but sharply incorrect in the Adverb Effect environment (see [213]): (212)

a. the vampire (that2) they said (*that1) attacked Tony b. a vampire that/*Ø under the right sort of conditions would certainly attack Tony

(213)

a. It was to Larry (that2) they said (*that1) were sent threatening postcards. b. It was to Larry that/*Ø at that time were sent threatening postcards.

Although in our terms here, the CT-arcs in both (212b) and (213b) should have finished 2-Quace and so not fall under (211), the Adverb Effect environment has no improving effect in cases like (212b) or (213b). I do not have an account of this fact. The only bright feature of the situation is that X-PP extraction and S3 extraction in the relevant contexts behave the same. To handle this in our terms here, it seems like it would be necessary to modify the Z-arc part of (198) so that it is only satisfied in a nonrelative clause. This ad hoc limitation on only part of a condition is evidently ugly and suspicious, but I have nothing better to offer at this point. 9.5. French Recall that section 8 showed that in French the analog of NEX clauses interacts with the French variant of the Anticomplementizer Constraint, that is, the obligatory mapping of complementizer que to qui in the case of S3 extractions, in such a way that, in contrast to English, the Anticomplementizer Constraint is not found in French NEX clauses. Combined with other factors, there is then no evidence that French analogs of X-PPs have any commonalities with French S3s at all. One approach to the facts would then just deny that there is quirky 1-Quace marking of French X-PPs. Another, which would keep the French construction abstractly more similar to the English one, would posit the same 1-Quace marking on X-PPs as has been posited for English but state CT-arc agreement in French, that is, the analog of (190), in a more restricted way, such that perhaps Quace agreement is only defined for non-Quirky 1-Quace, that is, for single-arc Quace Markers. I will not attempt to choose between such approaches here. It suffices that there is no particular mystery apparent in the French facts and no particular problem is raised for our approach to the English NEX clause phenomenon here. There is one further fact of relevance, though. This is represented by the English/French parallelism in (214):

82


(214)

a. the dragon (that) I believe (*that) was found b. le dragon que/*qui je crois *que/qui a été trouvé

Evidently, just as the English Anticomplementizer Constraint only holds for the minimal clause, the one that in this account contains the Start arc for the extraction, just so the French analog only holds there. This would indicate that parallel conditions are at work. In our terms here, it means that 1-Quace assignment to CT-arcs and additional assignment of 2-Quace to certain 30-arcs in contexts of ‘deep’ extraction are also found in French. 9.6. Final comment The key evidence that seemingly favors position (4c) is that X-PPs obey the various extraction constraints, the Anticomplementizer Constraint, the Parallelism Constraint, and the Complementizer Effect. I have tried to sketch, though, a method based on the notion of Quace that permits capturing the extension of these constraints, mostly relevant to S3 extraction, to X-PPs without the really radical and ultimately untenable idea that X-PPs are S3s. Rather, it can be stated that all these constraints pick out the relevant class of forms via specification of the notion of finished 1-Quace. And I further sketched how a parallel treatment was possible for the French analog of the Anticomplementizer Constraint. To conclude, one really terminological remark might be appropriate. The reader will have noted that in the development here, although it is denied that X-PPs are S3s, each X-PP does correspond to the head of a 1-arc—namely, the 1-arc type II local successor that provides its finished 1-Quace. Despite this, I think it is correct to avoid, as I have throughout, calling such phrases ‘subjects’. It is easy to give a general principle that underlies the choice of terminology (or nonterminology). Namely, a phrase has been called a ‘subject’ only if it heads a 1-arc of the sort that could be a superficial arc, that is, one that is not necessarily erased. Since every type II local successor is (by definition) erased, the inference is that merely heading a type II local successor R-arc is not grounds for being called an R. Adopted systematically then, this approach to terminology would not call a nominative object, one likely heading a final 2-arc with quirky 1-Quace, a subject and would avoid calling a dative subject, one, say, heading a final 1-arc with quirky 3-Quace, an indirect object. These terminological choices are formally clear, given the precise notion of type II local successor, and seem in accord with tradition as far as comparisons are possible. The only caution necessary then is that in an enriched arc framework that recognizes Quacearcs, it is simply not the case that, for example, merely being the head of an arbitrary 1-arc determines that a constituent is a subject, and so on.

2

A Putatively Banned Type of Raising

1. Background Syntacticians are a notoriously argumentative lot given to (sometimes) acrimonious disputes, and few, if any, syntactic conclusions have achieved uniform endorsement. One might therefore assume that a point on that a broad consensus of syntactic opinion has been reached is one that can be accorded some confidence. Notable then is that during the 1980s and 1990s proponents of the three most widely appealed to contemporary views of NL syntax, that is, the government binding (GB), lexicalfunctional grammar (LFG), and head-driven phrase structure grammar (HPSG) frameworks, all came to a common conclusion. Namely, although as a consequence of very different principles internal to each of these views, a certain type of syntactic raising was taken to be impossible. One might be tempted then to infer that this conclusion is now a well-established fact of NL syntax. In the framework of skeptical essays, however, caution is evidently in order. Postal (1974: 363, note 7) mentioned a class of English complement cases of the type shown in (1): (1)

You can depend on him to do something decent,

noting that they were likely part of a larger paradigm that also involves the verbs bet (on), bank (on), count (on), and rely (on). It was claimed that the highlighted DP in (1) was a superficial main clause subconstituent but an underlying complement subject, hence one raised into the main clause. A similar view was stated in Emonds 83

84


(1976: 77). Assuming that on + him in (1) forms a PP, a point to which I return, the suggested analysis would then instantiate what might be called raising to complement of P (RCP). Now precisely what the three views mentioned at the outset are agreed on is: (2)

RCP does not exist in NL.

If so, then the analysis of (1) must involve some distinct type of constructional features. However, the burden of this chapter is that RCP, rightly understood, is the factually supported analysis of a number of English cases like (1) and others as well. If so, it is necessary to gaze at no NL more exotic than English to determine that the agreed upon theoretical conclusion is wrong. The background of the question at issue in this chapter was nicely described in the following: (3)

McCloskey (1984: 441) “The relationship—syntactic and/or semantic—that holds between the matrix subject position and the null position marked by _ in an English example like (1) has been the source of much productive theorizing in the tradition of generative grammar: (1) Pascal appears _ to be playing well tonight It was in an attempt to elucidate the central properties of this relationship that the transformation of Subject (to Subject) Raising was first proposed (Rosenbaum 1967, Kiparsky & Kiparsky 1971, Postal 1974). It is with this ‘Raising relationship’ that I will be concerned in this paper, though in a different language—Modern Irish.”

McCloskey then focused the issue more sharply: (4)

McCloskey (1984: 441–442) “The dispute I will be concerned with is that concerning the range of positions to which a complement subject may legally be Raised. What is the range of possible targetpositions or host-positions for Raising? It seems to be uncontroversial that NP’s may be Raised to subject position. Raising to Object is widely believed to figure in the analysis of English examples such as those in (2): (2) a. The Press believes Mrs. Thatcher to be on the point of calling an election. b. I consider Spassky to have been the greatest player of his era. What I want to argue here is that the range of permissible options is in fact broader than this summary would suggest, that Raising to (a certain restricted subset of) prepositional object positions also exists. I will argue that the standard criteria for identifying Raising constructions suggest that there is a fairly large group of predicates in Irish that govern Raising, and that the Raised NP appears in a prepositional object position in the matrix clause.”

McCloskey had thus raised the issue of the viability of a ban on RCP almost twenty years ago but had done so on the basis of data external to English. In fact, as reiterated in Joseph (1990: 262), Joseph (1979) had still earlier attested an apparent case of RCP in Greek. Joseph (1990) argues in detail for the correctness of such an interpretation.

A PUTATIVELY BANNED TYPE OF RAISING

85

2. The LFG claim about RCP The earliest statement I am aware of that explicitly rejects the existence of RCP on theoretical grounds is part of the overall LFG view of Bresnan (1982b); see in particular the following (figure numbers have been changed to correspond to placement in this chapter): (5)

Bresnan (1982a: 348) “A further generalization of control follows from this theory. Recall that in lexically induced functional control relations, OBLq cannot be a controller. Since the oblique functions are marked by prepositions in English (Bresnan 1979), it follows that prepositional objects cannot be lexically induced functional controllers in English. The following examples were previously given to illustrate this point. (69) a. I presented it to John dead. b. *1 presented John with it dead. There are apparent counterexamples to this generalization, such as (70a,b): (70) a. Louise signaled to Ted to follow her. b. Mary relies on John to dress himself. “However, these are either cases of anaphoric control or cases of Verb-Preposition Incorporation (V-P Incorporation; see chapter 1). For example, the fact that we find Louise signaled to Ted for him to follow her, in which the subject is a lexically expressed NP, indicates that the complement of signal must be anaphorically, not functionally, controlled in (70a). As for rely on, it is subject to the rule of V-P Incorporation, which produces the two lexical forms shown in (71 [fig. 2.1]).

a.

b.

S

NP

Mary

relies

VP

NP

VP

PP

V

S

P

NP

on

John

Mary

V

NP

relies on

John

 2.1 Verb-Preposition Incorporation (71) a. [rely]V, (↑PRED) = ‘RELY-ON’ b. [rely on]V, (↑PRED) = ‘RELY-ON’ “The verb with lexical form (71a) is inserted into the structure shown in figure 2.1a; the verb with lexical form (71b) is inserted into the structure 2.1b. “In the structure shown in figure 2.1a, on John is a constituent, while in the structure shown in figure 2.1b, it is not. Moreover, John is an OBL in figure 2.1a, but an

86


OBJ in figure 2.1b. The former structure accounts for the impossibility of It is on John that Mary relies, in which the constituent on John is clefted; the latter structure accounts for John is relied on by Mary, in which John as an OBJ has passivized (chapter 1). The theory of control adopted here predicts that functional control of a complement to rely on should be possible only when the preposition is incorporated in the verb as in figure 2.1b. Hence, if the functionally controlled complement is present, on John must fail to form a constituent. This explains the contrast between (72a) and (72b) and the possibility of (72c). (72) a. *It is on John that Mary relies to dress himself. b. It is John that Mary relies on to dress himself. c. John is relied on by Mary to dress himself.”

In Bresnan’s LFG terms, ‘functional control’ is extensionally equivalent to the union of raising and control structures in most other syntactic approaches. The difference between raising and control is claimed to reduce to different functional identifications. So (Kaplan and Bresnan, 1982: 229): “The wider class of raising verbs differs from equi verbs in just this respect. Thus, the lexical entry for PERSUADE maps the baby f-structure in (108) into argument positions of both PERSUADE and GO. The OBJ of the raising verb expected, however, is an argument only of the complement’s predicate, as stipulated in the lexical entry (112).” The relevant claim of Bresnan’s (5) is then that the position of functional controller, that is, the position ordinarily taken to be the target of raising, cannot bear an Oblique relation. And since in English it is claimed that all DPs that are complements of PPs bear Oblique relations, the result that RCP cannot exist in English is predicted. With respect to a case like her (70b), Bresnan assumed that while it was a case of functional control of the nonraising type, this would not violate the stated constraints on the latter because such cases were claimed, on one analysis, to involve so-called V-P Incorporation, roughly, what is called reanalysis in other frameworks. According to this view, Bresnan’s (70b) would involve a reanalyzed structure in which there is no PP but rather a direct object. That is, in more standard terminology, Bresnan (1982a) was claiming that cases like her (70b) did involve raising of the complement subject into the main clause, as in the raising analysis of Postal (1974) for cases like (6): (6)

a. Arabella believed six to be a larger number than eight. b. Arabella proved six to be a larger number than eight.

The difference between Bresnan’s (70b) and (6) was, essentially, only that the former involved a verb that manifested V-P Incorporation. But in both types of case the raised phrase (functional controller) is a main clause object and not at any level of analysis internal to a PP. Moreover, Bresnan gave one argument for her V-P Incorporation view, namely, that it explained the badness of her (72a), since on the V-P Incorporation structure on + DP would not be a constituent and thus expectably not capable of being the focus of a cleft construction. However, this argument is undermined by a number of facts, which include several parallel to some invoked in Baltin and Postal (1996) as part of a general critique


87

of reanalysis. In particular, while Bresnan claimed that if the functionally controlled complement (the infinitive) was present, on + DP fail to form a constituent, one finds a range of examples like (7), which are inexplicable under that view: (7)

a. They were relying, it now appears/without any good reason/despite doubts, on the fifth division to cover their retreat. b. They were relying foolishly or at least naively on the fifth division to cover their retreat. c. They were relying on one division to cover their left flank and on another division to cover their right flank. d. They were relying not on the fifth division (to cover their retreat) but on the sixth division to cover their advance. e. One unit relied on the fifth division to cover the advance, the other on the sixth division to cover the retreat. f. They were relying on the fifth division or on the sixth division to come to their aid. g. Although they didn’t rely on the fifth division to cover their flank, they did on the sixth division. h. Napoli (1993: 363) On whom did you depend to make the arrangements?

It would clearly not be serious to regard the unique P on in (7a, b) as having incorporated into the verb and even less feasible to regard the (highlighted) second instances of P on in (7c, d, e, f, g) as having done so. Moreover, example (7h) indicates that for some speakers even the sort of extraction restrictions represented by Bresnan’s (72a), which she took to argue for a type of reanalysis, are not found (for at least some extractions). Moreover, observe the contrast in (8): (8)

a. b. c. d. e. f.

Ernestine had chastised Lewis during the meeting. Ernestine had chastised during the meeting—the guy who was opposed to her position. Ernestine had proved that guy to be an impostor. Ernestine had proved to be an impostor—the guy who was opposed to her position. Ernestine had counted/depended/relied on that guy to support you. *Ernestine had counted/depended/relied on to support you—the guy who was opposed to her position.

That is, the pivot in the clearer raising to object case (8c) can feed the Complex DP Shift phenomenon, like the uncontroversial object in (8a). But the pivot in (8e) cannot. If the pivot, for example, that guy in (8e), is inside a PP, the ill-formedness of (8f) follows from the independently motivated principle, known since Ross (1967/ 1986), that the Complex NP Shift phenomenon cannot strand prepositions. But if (8e) had a non-PP analysis of the sort entailed by Bresnan’s reanalysis/P-incorporation view and was really parallel to (8c), the ill-formedness of (8f) would remain unexplained. The facts gone over then mean that the grounds for the ill-formedness of Bresnan’s (72a) must be sought elsewhere.1 Overall then, Bresnan’s proposed uni-

88


versal constraint on functional control combines with her assumption that the cases in her (70b)/(72) involve functional control to just give the wrong answer for English, which provides no viable analysis for paradigms like her (70b)/(72), which include cases like (7) and (8e and f).2

3. The HPSG claim about RCP Both major statements of the HPSG framework so far, Pollard and Sag (1987, 1994), explicitly reject the existence of RCP: (9)

a. Pollard and Sag (1987: 20) “The existence of a syntactic dependency associated with raising explains the otherwise mysterious fact that there are no raising verbs with PP controllers analogous to PP-controlled equi verbs like appeal, for no PP can unify with the NP subject SUBCAT element of the verbal complement: (27) Kim appealed to Sandy to be optimistic. (PP object control)” b. Pollard and Sag (1994: 139) “In English, there is somewhat sparser evidence for the sharing of syntactic information in raising constructions. One suggestive piece of evidence, however, is the wellknown fact that in English PP objects may be equi controllers, but not raising controllers. That is, there are object equi verbs like appeal, whose infinitive complement’s subject is interpreted on the basis of the object of the prepositional phrase, as in (115): (115) Kim appealed to Sandy to cooperate. “But there are no analogous object raising verbs, that is, no verbs whose PP complement is not assigned a semantic role. We find no verbs otherwise like the raising verb believe that take PP complements whose prepositional object is a raising controller. Thus there are no verbs in English like the hypothetical kekieve in the following examples: (116) a. Kim kekieved to there to be some misunderstanding about these issues. b. Lee kekieved from it to bother Kim that they resigned.”

Where Bresnan’s LFG claim that barred RCP involved a constraint on so-called functional controllers, Pollard and Sag’s also purely theory-based rejection depends on the HPSG framework property that unification is not possible for a PP and the NP/DP subject SUBCAT element of verbal complements. This frameworkinternal theoretical difference aside, Pollard and Sag’s remarks add nothing to the evidence for the position that assumes the nonexistence of RCP attempted by Bresnan since they merely assert that the relevant cases do not exist. They made no attempt to analyze, for example, the potential counterexamples suggested in Postal (1974: 363, note 7), one of which was taken up by Bresnan. The lack of factual support offered is not obscured by their discourse, which refers to the putative nonexistence of RCP as a “mysterious fact,” a “well-known fact,” and even a piece of evidence for a theoretical conclusion. Since, unlike Bresnan (1982a), Pollard and Sag seem to grant that the raised phrases are parts of PPs and since they preclude an RCP analysis, their HPSG analysis would have to take the rely on


89

and similar cases as instances of control in the ordinary sense. I will return to the factual viability of such a view.3

4. The GB claim about RCP McCloskey (1984) specifically proposed to argue for the existence of RCP in Irish, against the background of the fact that this was clearly barred by GB framework assumptions, for reasons spelled out in detail as follows: (10)

a. McCloskey (1984: 444–445) “Chomsky (1981) argues for a certain view of the interaction between the notions of subcategorization, of thematic role assignment (q-role assignment) and the Projection Principle, which has the effect of radically restricting the range of variation in Raising constructions which the theory leads us to expect. One part of the theory of q-role assignment (the q-criterion) requires that no expression bear more than one q-role. Movement from one position which is assigned a q-role to another such position would result in the moved category being assigned two q-roles—one from its original position, and one from the position to which it is moved. Therefore movement from a q-position (a position assigned a q-role) is possible only if the target of the movement is not a q-position. It is assumed moreover, within the GB framework, that subcategorization implies q-marking—that is, that a verb assigns a q-role to each position in VP for which it is subcategorized. It follows, of course, that movement to a subcategorized position will always give rise to a violation of the q-criterion and is thus illegitimate. More generally, there can be no movement to any position within VP (Chomsky 1981, p. 46), since other assumptions within the framework in question imply that movement to non-subcategorized positions within VP would also give rise to violations of the q-criterion.” b. McCloskey (1984: 445) “This combination of assumptions is linked to a much more general principle—the Projection Principle—with broader consequences. The Projection Principle holds that all levels of ‘syntactic’ representation (D-structure, S-structure and LF) are isomorphic and reflect directly the thematic and subcategorizational properties of lexical items. Raising to a position within VP, would violate this principle, because Raising verbs take only clausal complements at LF as a lexical property, but analyses which include Raising to a position within VP need to provide these verbs with an extra complement NP at D-structure (where it would be empty) and at S-structure (where it would be filled by the Raised NP). Such analyses then imply a mismatch within VP between the properties of this group of verbs at LF on the one hand, and at D- and S-structure on the other.”

Having stated clearly how and why the GB system precluded inter alia RCP, McCloskey went on to argue that Irish in fact allowed RCP in certain cases, a claim challenged in Stowell (1989). I will not be concerned with this Irish-internal issue here (see Postal, 1986b). The argument here is that any collection of principles that

90


bar RCP can be shown to be incorrect merely on the basis of English facts, moreover, facts that are rather different in form from those McCloskey and Stowell treated.4

5. The English class: bank/bet/count/depend/rely + on + infinitive 5.1. Remarks My goal is thus to argue that RCP is instantiated in English. As in Postal (1974), I refer to the DP whose status is at stake in the issue of whether a particular case involves raising or control (or something distinct) as the pivot DP. In the cases here, the pivot is then a PP object. The clearest evidence of English RCP involves the verbs listed in the heading of this section, hereafter the on verbs.5 The reasons are as follows. First, as already touched on in section 2, the relevant class of infinitival structures actually involves PPs. These are fairly evidently main clause constituents. So consider: (11)

a. One can count on the ex-president with a high degree of confidence to lie on most public occasions. b. They were depending on the monster, apparently, to rush back into the cave. c. Most staff members were relying on the director in a touchingly naive way to do the right thing.

In each case, the highlighted modifier is (can be) understood to modify the predicate of the main clause. This makes it implausible in the extreme to regard, for example, on the director in a touchingly naive way to do the right thing to be a complement clause or even a constituent. This conclusion is supported by the fact that such sequences cannot be the foci of pseudocleft or Right Node Raising constructions: (12)

a. *What most staff members were counting was on the director (in a touchingly naive way) to do the right thing. b. *Many managers may have been relying and some workers were certainly relying— on the director (in a touchingly naive way) to do the right thing.

Second, the PP complement in on verb cases like (11) is evidently understood as the subject of the infinitival constituent. There are then two standard conceptualizations of such facts. One can either invoke some kind of raising or posit a control relation between an unraised main clause (PP) object constituent and the complement subject, taken, for example, to be some sort of invisible element. In effect, under the raising view, the putatively raised phrase is a syntactic but not a semantic constituent of the main clause, while under the control from main clause object view it is both. Acceptance of the raising view in the on verb case of course grants the claim of this investigation. Therefore, it seems that to reject that, one must adopt a control view of cases like (11), as implied in the Pollard and Sag (1987, 1994) materials quoted earlier. To justify this position, it then largely suffices to show that a control analysis of on verb structures is not viable.


91

5.2. Traditional raising/control differentia There are of course a number of traditional criteria that argue that a pivot is not a controller and thus, if one accepts that the only alternative is a raising analysis, that it is a raisee. These criteria hold of the least controversial instances of raising, those where the pivot ends up as the subject of a nonpassive, as with verbs like appear/ seem/turn out or adjectives like bound/certain/likely, and so on. The first relevant property is that a controller is normally involved with selections in the main clause, as well as in the complement. Lack of any selectional restrictions with respect to the main clause is then an argument against controller status. Notably then, the PP objects of on verbs at issue do not seem to be selected in the main clause and are extremely free. They can be animate or inanimate, human or nonhuman, count or mass, abstract or concrete: (13)

a. b. c. d. e. f. g. h.

One can’t depend on that gorilla to behave himself. I depended on my car to start in the winter. I depended on the nurse to watch over my uncle. I depended on the alarm to go off at the right time. I depended on the carrots to peel rapidly. I depended on the soup to cook properly. I depended on her understanding to be sufficiently sharp for the job. Michael relied on that algorithm to properly sort people into friends and enemies.

This situation contrasts with that of an uncontroversial object control type pivot such as the object of, for example, convince, which must, modulo personification, of course be animate: (14)

a. b. c. d. e. f. g. h.

One can’t convince that gorilla to behave himself. *I convinced my car to start in the winter. I convinced the nurse to watch over my uncle. *I convinced the alarm to go off at the right time. *I convinced the carrots to peel rapidly. *I convinced the soup to cook properly. *I convinced her understanding to be sufficiently sharp for the job. *Michael convinced that algorithm to properly sort people into friends and enemies.

The contrasts between (13) and (14) roughly parallel those between the uncontroversial raising structures with seem/likely and the uncontroversial control structures with want/hope, and so on. A second criterion, related to the selectional facts just cited, is that control structures are in general incompatible with the possibility of a pivot being an expletive. In uncontroversial raising and control cases, this yields well-known contrasts like the following: (15)

a. There are likely to be spies in the class. b. *There want to be spies in the class.

92


(16)

a. It is likely to be evident that the bomb will not be found. b. *It wants to be evident that the bomb will not be found.

(17)

a. It is likely to sleet. b. *It wants to sleet.

(18)

a. It is likely to be astounding, the number of people who choose to eat snails. b. *It wants to be astounding, the number of people who choose to eat snails.

Significantly then, many, though not all, expletives are possible as the PP objects with on verbs, but none are possible with, for example, convince: (19)

a. b. c. d.

*Amanda counted/depended/relied on there to be spies in the class. Amanda relied on it to be evident that the bomb would not be found. Amanda depended/relied on it not to sleet. Amanda banked on it to be astounding, the number of people who chose to eat snails.

(20)

a. b. c. d.

*Amanda convinced there to be spies in the class. *Amanda convinced it to be evident that bomb would not be found. *Amanda convinced it not to sleet. *Amanda convinced it to be astounding, the number of people who chose to eat snails.

This evidence, too, supports the noncontroller status of the PP objects; the fact that expletive there is not always possible where ‘expected’ does not really undermine this conclusion, since it is easily seen that there are a host of special constraints on there even in clear noncontrol and probable raising cases, such as (21–23): (21)

a. Herbert made it rain/be possible to travel to other galaxies. b. *Herbert made there be a riot.

(22)

a. Herbert watched it rain/get dark/become impossible to get a drink after 11 P.M. b. *Herbert watched there be an investigation of his father.

(23)

a. Herbert considered it likely to be impossible to square that circle. b. *Herbert considered there likely to be a runoff election.

Moreover, even the more restricted expletive there is sometimes possible in on verb infinitives: (24)

a. Don’t count on there to be that many supporters in the organization. b. She should not have depended on there to be as much resistance as she hoped. c. It is never a good idea to rely on there to be more favorable opinions than unfavorable ones. d. We can’t depend on there to be enough beer to keep all the students happy.


93

I am not sure about what picks out that subclass of on verb environments that allow expletive there, but it seems to have something to do with the complement that represents a specification of quantity. A third traditionally cited argument against the controller status of a pivot position (and hence in favor of its raisee status) is that raising structures fairly freely allow restricted DPs, for example, idiom chunks, whereas the control cases mostly do not.6 So there are well-known idiom chunk contrasts like the following: (25)

a. b. c. d.

Close tabs were kept/*placed/*maintained on her movements. They believe close tabs to have been kept/*placed/*maintained on her movements. *They convinced close tabs to be kept/placed/maintained on her movements. ?They were counting/depending/relying on close tabs to be kept/*placed/maintained on her movements. e. ?Close tabs were being counted/depended/relied on to be kept/*placed/*maintained on her movements.

(26)

a. The shit hit/*struck/*collided with the fan. (The * notation here indicates lack of idiomatic meaning = ‘something bad happen’.) b. They believe the shit to have hit the fan yesterday. c. *They convinced the shit to hit the fan yesterday. d. They were counting/depending/relying on the shit to hit the fan when the new director arrived. e. The shit can be counted/depended/relied on to hit the fan at that time.

(27)

a. That finally broke/*crushed/*smashed the ice at the meeting (* = lack of idiomatic meaning = “initial interpersonal tensions were dissolved”). b. The ice was finally broken/*crushed/*smashed. c. *They convinced the ice to finally be broken. d. They were counting/depending/relying on the ice to finally be broken. e. ?The ice can’t be counted/depended/relied on to finally be broken.

(28)

a. Her ship/*boat finally came/*floated in (= “she received the life-improving lucky break”). b. Her ship is likely to come/*float in. c. They believed her ship to have finally come/*floated in. d. You can’t count/depend/rely on your ship to come/*float in just when you want it to.

(29)

a. b. c. d.

(30)

a. *Her ship wants to come/float in. b. *They convinced her ship to come in.

The lovebug bit/*clawed/*grabbed Marsha (= “Marsha fell in love”). The lovebug seems to have bitten/*clawed/*grabbed Marsha. It is a mistake to believe the lovebug to have bitten/*clawed/*grabbed Marsha. It is a mistake to count/depend/rely on the lovebug to bite/*claw/*grab Marsha in time.

94


(31)

a. *The lovebug wants to bite Marsha. b. *They convinced the lovebug to bite Marsha.

(32)

a. b. c. d.

All hell broke loose/*broke up/*broke down. All hell seems to have broken loose/*up/*down. It is a mistake to believe all hell to have broken loose/*up/*down. You can count/depend/rely on all hell to break loose/*up/*down at that time.

That is, the on verb + PP object + infinitive cases behave like the uncontroversial raising cases with seem/likely/believe, not like uncontroversial control cases. 5.3. Nontraditional raising/control differentia I believe several infinitival characteristics not standardly cited, systematically differentiate raising and control structures. One involves the property originally taken in Postal (1970) to be characteristic of control structures, namely, that the controlled complement has an invisible (weak definite) pronominal subject.7 This then theoretically differentiates such complements from those of raising structures in which, in the general case, the complement subject simply is the main clause pivot DP, which, of course, need not be a pronoun.8 Therefore, this theoretical difference would in principle permit factual differentiation of the two structures in specific cases if one could find independent grammatical features that reveal the presence or absence of (invisible) weak definite pronouns. A feature of this type that I have made much of in other contexts involves what I have called antipronominal contexts, those that reject the presence of weak definite pronouns.9 If one finds a context, in this case, in particular, an antipronominal subject context, then the theoretical distinction between raising and control complements in the terms of Postal (1970) would lead one to expect (33): (33)

Whereas, modulo note 7, a raising complement should be indifferent to whether its subject position is antipronominal, a control structure should be ungrammatical when the complement subject position is antipronominal.

Unfortunately, I have not found many subject positions that are antipronominal, which renders the task of testing (33) rather difficult. Still there are some. Two include the subject positions of be the matter/wrong with: (34)

a. Something/That/Lots of things are the matter with my transmission. b. Something1 is the matter with my transmission, but that sort of thing/*it1 is not the matter with his. c. That kind of thing/Nothing is wrong with her values. d. *He said something1 was wrong with her values and it1 is wrong with them.

Given that the subject positions of be the matter with and be wrong with appear to be antipronominal, one can test (33) with uncontroversial raising and control cases.


95

Significantly then, one observes first that such contexts do not combine with clear control structures: (35)

a. (Its) Being common doesn’t mean that that problem/something/anything is serious. b. *(Its) Being the matter with my transmission doesn’t mean that something/anything is serious. c. *Something/Lots of things can be the matter with your transmission without being the matter with mine.

(36)

a. (Its) Being harmful may indicate that something is detectable. b. *(Its) Being wrong with your liver may not mean something is detectable. c. *That can be detectable without being wrong with your liver.

And equally clearly, they do combine with a variety of clear raising structures: (37)

a. Something/Lots of things/Nothing seem(s) to be the matter with your transmission. b. Many things are likely to be the matter with your transmission. c. What they proved to be the matter with your transmission is not serious.

(38)

a. Did that appear to be wrong with Mike’s liver? b. Such a thing is bound to be wrong with someone’s liver. c. They didn’t believe anything to be the matter with her liver.

Against this background, it is thus possible to test whether the behavior of the on verb structures resembles those of clear control cases or those of clear raising cases: (39)

a. You can count on that not to be the matter with your transmission. b. He wasn’t depending on anything of the sort to be the matter with the enemy’s missile defense system. c. We can’t rely on that many things/all those things to be wrong with their intelligence network.

The evidence then shows that the infinitival complements of on verb structures behave like the uncontroversial raising cases and contrast with the uncontroversial control forms. A second nontraditional way of testing the difference between raising and control cases relates to the possibility of the complement representing a middle structure, which, I suggest, is subject to condition (40): (40)

A control complement cannot be a middle clause.

It is apparent that this divides up uncontroversial cases properly: (41)

a. Harry bribed Uruguayan police. b. Uruguayan police bribe cheaply/easily.

96


c. d. e. f. g. h. i.

Uruguayan police seem/are likely to bribe cheaply/easily. *Uruguayan police hope/try/want to bribe cheaply/easily. *Bribing cheaply/easily pleases Uruguayan police. Hugh proved Uruguayan police to bribe easily/cheaply. *Hugh persuaded Uruguayan police to bribe easily/cheaply. *Some policemen can be poor without bribing easily. *Uruguayan policemen often apologize to foreigners for not bribing cheaply.

(42)

a. b. c. d. e. f. g. h.

They frightened the wimp. Certain wimps frighten easily. Certain wimps seem/are likely to frighten easily. *Certain wimps don’t try/want to frighten easily. *Frightening easily worries certain wimps. They proved certain wimps to frighten easily. *They persuaded certain wimps to frighten easily. *Good soldiers can be inexperienced without frightening easily.

(43)

a. b. c. d.

They sold the beer easily. Such beer sells easily. *Selling easily made that sort of beer expensive. *That kind of beer can be cheap without selling easily.

Note, too, clear correlations like (44): (44)

a. b. c. d.

They prevented there from being a riot. They prevented Uruguayan police from bribing cheaply. *They discouraged there from being a riot. *They discouraged Uruguayan police from bribing cheaply.

That is, traditionally prevent is analyzed as inducing raising (to object) (see Postal, 1974: 154–163; and Postal and Pullum, 1988: 655–657), supported by the expletive object of (44a), which analysis is then consistent with (40) given the grammatical middle complement of (44b). On the contrary, discourage, which as (44c) illustrates, precludes expletive objects, and is traditionally taken to induce control, allows no middle complement, as (44d) indicates. Significantly then, the infinitival complements of on verbs can be middles: (45)

a. Hugh was counting/depending/relying on Uruguayan police to bribe easily/cheaply. b. They counted on/depended on/relied on certain wimps to frighten easily. c. She was counting on that sort of beer to sell easily.

These data thus also support the view that on verb structures are of the raising, not the control, type. The argument from middle distribution in favor of a raising analysis of on verb infinitival structures can be a bit strengthened as follows. The facts with the on verb instances of middle infinitives with pivot DPs inside PPs contrast with other pivot


97

instances inside PPs, that is, with real control cases, which are ungrammatical, as (40) specifies: (46)

a. *The president appealed to/proposed to/suggested to Uruguayan police to bribe cheaply. b. *The president pleaded with Uruguayan police to bribe cheaply.

One potential problem with the argument for a raising analysis of on verb cases based on principle (40) has been pointed out to me by Idan Landau (personal communication of July 16, 2000). He observed that the infinitive in the case of the infinitival complements of the main verbs claim and pretend can be a middle, providing examples (47a and b), to which I add others: (47)

a. b. c. d.

Linguists usually claim to discourage easily. Many linguists pretend to discourage easily. Uruguayan police claim not to bribe cheaply. Uruguayan police pretend to bribe cheaply.

I agree that these are fine. If, as usually assumed, these are control cases, they obviously falsify (40), raising then the issue of why the other control cases are bad. Idan Landau offered the suggestion that the right generalization takes account of the fact that (47a and b), unlike typical control cases, involve complements that represent propositions. One must then consider an alternative to hypothesis (40) of the following form: (48) An infinitival or gerundive complement that is nonpropositional cannot be a middle.

Differentiating the factual viability of this generalization, which would not help differentiate raising and control structures, from that in (40) is not so easy because most infinitives are nonpropositional. Still there are a few pretty clearly propositional ones: (49)

a. Sheldon demonstrated/proved to us to know Spanish (= that he knew Spanish). b. Gwen swore to us to be a government agent (= that she was a government agent).

These structures are not predicted by (48) to bar middles, any more than (47a and b) do, while (40) claims that middles should be impossible. The latter is correct: (50)

a. *Corrupt police try to demonstrate to their clients to bribe cheaply. b. *Even cowards may swear to TV reporters not to frighten easily.

Further, when one looks at gerundive clauses that represent propositions, middles are also impossible as (40) specifies, while (48) would allow them to be good: (51)

a. Being a vampire (= ‘that she is a vampire’ = ‘that it is true that she is a vampire’) proves Jane not to be a vegetarian. b. *Bribing cheaply proves Uruguayan police to be desperate.

98


c. *Frightening easily makes wimps unreliable. d. *Rattling so easily makes a TV announcer appear foolish.

Moreover, observe that middles are impossible in clearly proposition-denoting controlled adjuncts: (52)

a. Uruguayan police can be corrupt (without looking guilty/*without bribing easily). b. Uruguayan police were criticized for being dishonest/*bribing too cheaply.

My suggestion is thus that one should accept the incompatibility of middles and control structures and therefore take the grammaticality of (47) to falsify not (33) but rather the traditional view that such cases are control structures. I propose instead that they are raising structures associated with the additional feature (quite exceptional in English) that the raised form, understood as bound by the main clause subject, is invisible.10 It is not irrelevant, I think, that these verbs do take infinitival complements transparently not interpretable as control cases but easily seen as raising ones, less marginally so in the case of claim: (53)

a. b. c. d. e. f. g. h. i.

*She claimed that to be the case. *They claimed biological weapons to be too terrible to use. *Irving pretended something to be bothering him. That, he claimed to be the case. Biological weapons, they claimed to be too terrible to use. What did they claim to be too terrible to use? What Irving pretended to be bothering him was his neighbor’s drum playing. No matter what they pretend to be impossible, don’t believe them. That, he may have only pretended to be the case.

While straightforward cases like (53a, b, and c) are obviously ungrammatical, they arguably manifest what was called in Postal (1974: 305) the Derived Object Constraint (DOC). This bars certain raised objects from remaining in the standard main clause postverbal object position. But cases like (53d, e, f, g, h, and i), which satisfy DOC, seem fine. Thus a consistent analysis can claim that these two verbs, not mentioned in Postal (1974), are also subject to DOC. Consider then: (54)

a. *Marilyn claimed herself to be intuitive. b. *Marilyn pretended herself to be intuitive.

These would surely be regularly ruled out by any formulation of DOC capable of blocking (53a, b, c). Suppose, though, continuing to assume these are raising cases, that these two verbs allow, and arguably in fact require, raised reflexive objects bound by their main clause subjects to be invisible.11 This would properly differentiate the following: (55)

a. *Herself1, Marilyn1 claimed to be intuitive. b. Herself2, Marilyn1 claimed to have talked to Jane2 about.


99

c. *Herself1, Marilyn1 pretended to be intuitive. d. Herself2, Marilyn1 pretended to have talked to Jane2 about.

A further bit of evidence that the infinitival complements of the verb claim are raising, not control, structures is that they permit the same type of passives as less controversial raising inducers such as believe and prove: (56)

a. b. c. d. e. f.

The gorilla was claimed/believed by Austin to be telepathic. Such things are often claimed/proved by scientists to be dangerous. That program was claimed to have been written in Cobol. It was claimed to have been hidden from the people that he was a werewolf. *It was pretended to be the case. *That was pretended to be impossible.

There is clearly no way to regard such claim cases, which, it should be observed, satisfy DOC, as control structures, since uncontroversial instances of the latter cannot correspond to such passives: (57)

a. Herb hopes/longs/tries to be brave/help the handicapped. b. *Those people are hoped/longed/tried by Karen to be brave

Left open of course is why the putative raising complement of pretend does not permit passivization, a question not directly relevant to our concerns here. A virtue of an obligatory reflexive deletion view of what I am arguing are raising cases like (53d, e, f, g, and h) is that it is consistent with the fact that such cases are understood de se, a systematic feature of true control structures.12 This might seem to argue against a raising analysis but does not if the latter is limited, as here suggested for these cases, to reflexive structures, which also are well known to have the de se requirement; see Higginbotham (1992: 87). That is, with respect to that feature, the following pairs do not differ: (58)

a. b. c. d.

Glen expects himself to win the first match. Glen expects to win the first match.13 Glenda wants herself to develop better muscle tone. Glenda wants to develop better muscle tone.

A third poorly known property that distinguishes control from raising cases involves certain metonymous stock (price) structures: (59) Microsoft went up.

In this case, a company name DP appears as a subject with certain predicates but is understood to refer to the stock of that company, in turn to the price of the stock of that company. So (59) is equivalent to (60a), in turn equivalent to (60b): (60)

a. Microsoft stock went up/down/dropped/rose (in price). b. The price of Microsoft stock went up/down/dropped/rose.

100


Now, relevantly, the usage in (59) does not interact freely with complement types and for uncontroversial cases, a generalization initially statable as (61) holds: (61)

The ‘stock’ usage of company name DPs like that in (59) is not possible in control structures.

This generalization is supported by data like (62): (62)

a. b. c. d. e. f. g. h. i.

Microsoft seems/is likely to have gone up. They proved Microsoft to have gone up. *Microsoft hopes/plans/tried/wants to go up. *Going up pleased Microsoft. *They persuaded Microsoft to go up. Microsoft hired Michael Downs without knowing that/*going up. Microsoft hired Michael Downs to impress the government/*go up. *Microsoft fired its chairman after going down. *Microsoft denied the story without shooting up/rising much.

And here also the usage in question is perfectly possible in the infinitival complements of on verbs: (63)

a. The labor union was counting on Microsoft to go up toward the end of the year. b. No one should depend on Enron to go down/drop any further. c. I can’t rely on Enron to rise this year.

So constraint (61) also argues for the raising and against a control analysis of the infinitives with on verbs. It is important to dispel a possible objection to (61). Despite ill-formedness like that in (62c–i), control cases like (64) are perfectly grammatical: (64)

a. Microsoft went up today after going down yesterday. b. Going up yesterday did not cause Microsoft to go down today.

This shows that the initial extremely informal formulation in (61) is inadequate. However, I do not think such cases cast doubt on the reality of a control restriction on the relevant stock usage. For it is easy to see that cases like (64) differ in a specific way from the ungrammatical control structures taken to support (61). Namely, in the latter the controller company name DP as it were stands alone in a position where in general a DP that denotes stock or its price would be ungrammatical. That is, it appears in a position that selects mind-possessor denoting DPs, with companies being taken to be of that category. In (64), however, the controller is a company name DP that is itself understood to denote the stock or stock price of the relevant company. My partially speculative suggestion is that one can give a superior replacement of (61) that allows (64a, b) as follows. Assume that the stock (price) interpretation of a company name DP simply involves a specific type of possessor ascension so that there is a structure for cases like (64), in which the DP Microsoft is the possessor of


101

a larger DP headed by a noun that means “stock” and that that larger DP is in turn the possessor of one whose head noun means “price.” Schematically then, (65a) would have a representation like (65b): (65)

a. Microsoft went up. b. Microsoft’s1 stock’s2 price3 went up.

Now, under this three DP analysis, grammatical cases like (64a and b) are represented as (66a and b), respectively, while an ungrammatical one like (62c) is represented as (66c): (66)

a. Microsoft’s1 stock’s2 price3 went up today after [it]3 going down yesterday. b. [It’s]3 Going up yesterday did not cause Microsoft’s1 stock’s2 price3 to go down today. c. *Microsoft1 wants its1 stock’s2 price3 to go up.

The key point then is that under this speculative analysis, in the ungrammatical (62c) = (66c), the controller is a company denoting DP, while in the grammatical (64a, b) = (66a, b) the controller is not a company denoting DP but one that denotes price. This suggests that the right elaboration of (61) would be something like (67): (67)

A proper noun-headed company-denoting DP cannot be the antecedent (controller or not) of a pronoun that is a (possessor ascended) company denoting DP.

I have generalized (67) beyond controllers to cover as well facts like (68): (68)

a. *Microsoft1 hired Bevins because it1 had gone down recently. b. *Microsoft1 claimed that it1 would go up. c. *He wants to work for Microsoft1 because it1 will soon go up.

Compare: (69)

Microsoft went down recently because it had gone up so much before.

As with the control cases (64a and b), (69) is good because the antecedent of the pronoun can be taken to be a larger DP that contains the company name phrase, one that denotes stock price. Inter alia, (67) accounts for the lack of ambiguity of the long form of (70), whereas the short form is ambiguous: (70)

Jerome bought Microsoft1 (because it1 went down).

That is, the short version has readings where what is bought is the company or the stock of the company, but the long version has only the stock reading. Another virtue of the formulation in (67) that covers pronouns and reflexives is that it can account for the fact that the stock usage in question does not seem acceptable with the verbs claim and pretend, argued earlier to involve raising not control:

102 (71)


a. *Microsoft claims to have gone up. b. *Microsoft pretended to drop.

Given the earlier proposal that claim/pretend infinitival structures involve invisible raised reflexives bound by the subject, what is wrong with (71a and b) is that they violate the same constraint seen in (72): (72)

a. *Microsoft1 claimed that that was the case after it1 went up. b. *Harry wants to work for Microsoft1 because it1 is going to go way up. c. *Microsoft believes itself to have gone up.

That is, very roughly, it does not seem possible for a pronominal form, reflexive or not, that designates the stock or price of the stock of a company to be anteceded by a DP that designates that company. That is, given the formulation in (67), cases like (72) are blocked under either a control or a raising analysis. Just as was remarked apropos of constraint (40), support for (67) is also increased when it is observed that clear instances of control that involve pivots internal to PPs contrast with on verb cases like (63): (73)

a. *The president appealed to/proposed to/suggested to Microsoft to go up. b. *The president pleaded with Microsoft to go up.

It must be stressed that the English grammatical constraint here reconstructed as (67) can not be understood in general semantic terms. What I mean can be indicated with respect to (74): (74)

*Microsoft intends to go up.

It might occur to some to suggest that what is wrong with such a case is that there is a semantic clash between the controller, a DP that denotes a company, and the controlee, which designates a stock or its price. But it can be shown beyond serious doubt that this is not the relevant factor. This is possible because a variety of other metonymy cases with the same semantic clash property characteristic of the stock cases are grammatical. Here are two: (75)

a. I am parked on Twenty-sixth Street (= “my car is parked on Twenty-sixth Street”). b. I didn’t want/intend to be parked on Twenty-sixth Street.

(76)

a. That pitcher is hard to hit (= “That baseball pitcher’s pitches are hard to hit”). b. That pitcher wants/intends to be hard to hit.

In (75b) the controller denotes a person, but the controlled form denotes a vehicle; in (76b) the controller also denotes a person, but the controlled form denotes (baseball) pitches. Hence it is clear that there is no general principle that requires some sort of semantic matching that could fail in the stock case. Consequently, one must appeal to a construction-specific constraint for the latter such as (67).


103

A fourth property that distinguishes uncontroversial control and raising structures from each other but is not part of the traditional battery of tests for the distinction is called partial control in Landau (2000, 2002). This is the phenomenon seen in cases like the following: (77)

a. *The director met at noon. b. The director wants to meet at noon.

(78)

a. *I got together yesterday. b. I wanted to get together yesterday.

That is, with intransitive verbs that normally require a plural or conjoined subject, like meet, get together, join together, divorce, and so on, it is possible to find cases where they occur as the heads of infinitival complements of what are universally analyzed as control structures where the controller is a singular. Taking these complements to have invisible subjects, the interpretation then is that the denotation of the controller is understood to be (only) one element of the set denoted by the invisible subject, the others being unspecified. The relevance of this phenomenon, treated in detail in Landau (2000), is that, as he states clearly, it is impossible in uncontroversial raising cases: (79)

a. b. c. d. e.

*Myron1 seems [ ]1, . . . n to have met at noon. *Myron1 is likely [ ]1 . . . n to get together tomorrow. *I believe Myron1 [ ]1, . . . n to have met at noon. *They proved Myron1 [ ]1, . . . n to have gotten together at that time. *They prevented Myron1 from [ ]1, . . . n getting together at that time.

Testing then how the PP structures that occur with on verb infinitivals behave with respect to partial control, one finds that they fall together with the uncontroversial raising cases and contrast with control cases: (80)

a. *I can’t count on Myron1 [ ] 1, . . . n to get together when we want to. b. Myron1 refused [ ] 1, . . . n to get together when I wanted to.

(81)

a. *One shouldn’t depend on Myron1 [ ] 1, . . . n to meet at all. b. Myron1 hoped [ ] 1, . . . n to meet before the demonstration.

(82)

a. *They were relying on Myron1 [ ] 1, . . . n to get together after the party. b. Myron1 worried about [ ] 1, . . . n getting together without her.

Here also then the evidence supports the raising character of on verb infinitival structures, not a control analysis.14 As with the earlier arguments, this support for a raising analysis of on verb infinitives can be broadened by again noting the contrast with clear control cases that have pivots internal to PPs. So compare (80)–(82) with (83):

104 (83)


a. Carla appealed to/proposed to/suggested to Larry1 [ ] 1, . . . n (not) to meet without her. b. Carla pleaded with Larry1 [ ] 1, . . . n not to meet without her.

An evident implication of the combination of the view that partial control is impossible in raising structures with my claims that the verbs claim and pretend that occur with infinitival complements represent raising, not control, structures is that partial control should be impossible in forms like, for example, (84): (84)

a. *The department chairman1 claims [ ] 1, . . . n to have gotten together/met yesterday. b. *The department chairman1 pretended [ ] 1, . . . n to meet yesterday.

I find these to indeed be ungrammatical. However, Landau (2002) cites the following as grammatical: (85)

The chair1 claimed [ ] 1, . . . n to be gathering once a week.

But I find (85), if anything, even worse than (84a). One way, perhaps, to sharpen the issue is to note that the control view, which would take (85) to be good, predicts a clear difference in such sets as (86): (86)

a. *The department chairman1 claimed [ ]1, …n to have gotten together last week. b. *The department chairman1 was claimed [ ]1, …n to have gotten together last week. c. *The department chairman1, they claimed [ ]1, …n to have gotten together last week.

Since it is clear that the latter two examples cannot involve control but must be raising structures, for me all three should be undifferentiated, which is how I judge them. 5.4. A problematic type of restriction While the evidence cited so far seems to fairly relentlessly support the raising nature of on verb infinitival complements and disconfirm any view that they involve control structures, one set of facts might seem to raise difficulties. It was suggested in Rosenbaum (1967) that an effective test for the raising/control distinction is the way the two types of cases interact semantically with passivization. When all relevant DPs are definite referential expressions, active and passive are semantically (truth functionally) equivalent: (87)

a. Glen tickled Betty. b. Betty was tickled by Glen.

That is, one of these can represent a true assertion if and only if the other does. When such passives interact with clear raising/control structures, though, equivalence is found only in the former: (88)

a. Glen seems to have defended Betty = b. Betty seems to have been defended by Glen = c. It seems that Glen defended Betty.


105

d. Glen wants to defend Betty ≠ e. Betty wants to be defended by Glen.

The same story holds for the object cases like (89): (89)

a. b. c. d.

Stan expects Glen to defend Betty = Stan expects Betty to be defended by Glen. Stan persuaded Glen to defend Betty ≠ Stan persuaded Betty to be defended by Glen.

The equivalence between (88a and b) would follow from the simple passive equivalence if the meaning of the (88a) case were that of the full that clause case, that is, if the pivot NP in the (88a) case with seem played no semantic role. By parity of reasoning then, the structure of the (88d) case should be different and in particular should be such that the pivot NP does play a semantic role in the main clause. For then the nonequivalent passives will involve two different NPs in the main clause, and that is how they differ semantically. That is, in (88c) the sentence is about a want of Glen’s but in (88d) about a want of Betty’s. According to this reasoning then, equivalence under passivization is a test for raising structures. Attempting to apply this idea to on verb cases, one finds, however: (90)

a. Stan was counting/depending/relying on Glen to defend Betty. b. ?*Stan was counting/depending/relying on Betty to be defended by Glen.

The first point is that whereas (90a) are normal sentences, the sentences in (90b) seem unacceptable. Second, cutting through the obscurity determined by its ungrammatical status, (90b) does not seem equivalent to the corresponding case of (90a). Roughly, the latter seems to indicate that the focus of the ‘reliance’ is on the entity denoted by the raised DP Betty, while the former indicates that the focus is on the entity denoted by the raised DP Glen. This state of affairs is not consistent with what Rosenbaum dealt with. Now, encounters with instances of this problem are of long date: (91)

Postal (1974: 363–364) “An interesting set of cases relevant to assumption linkage involves a set of verbal elements with the form Verb + on, including bet on, depend on, bank on, count on, rely on. Consider depend on. This occurs with both gerundive and infinitival complements: (i) a. You can depend on him to do something decent. b. You can depend on him doing something decent. his “I should like to argue that sentences like (i)a are, in fact, derived by Raising. However, examples like (i)a and (i)b are not strict paraphrases. Compare the following: (ii) a. ?You can depend on that corpse to remain here for another hour. b. You can depend on that corpse’s remaining here for another hour. “It seems clear again that the putative Raising examples entail the analogous unraised examples, but not conversely. That is, there seems to be an assumption linked with

106


Raising application for depend on. Roughly this is that the dependable state of affairs is so because of the will(s) of the entities designated by the raised NPs. Hence (ii)a is anomalous because corpses have no wills, etc. Since it is easy to show that the post-on NP with infinitival complements in such cases is a main clause constituent, the alternative to a Raising analysis would involve distinct underlying structure types for the a and b examples in such pairs, a highly unsatisfactory account. I think that these verbal examples with on should thus provide a fruitful domain for studying the interaction between Raising application and particular linked semantic assumptions.”

These remarks were made internal to a section that discussed a number of cases that putatively involved raising being associated with particular semantic assumptions. I cannot claim to understand much more about these matters than when I wrote the preceding. However, if it is accepted that such linkages of raising with implicatures are a real phenomenon, then the facts in (90) would not really bear on the fundamental issue of these remarks, which is the existence of RCP and its support from on verb structures. Significantly then, I argued a bit in the section cited here that even uncontroversial raising structures can reveal the kind of linked semantic assumption seemingly found with on verbs that take infinitives. So compare the following: (92)

a. Joseph Stalin seemed to me to be an evil monster. b. It seemed to me that Joseph Stalin was an evil monster.

For someone like me, who had the good fortune to never encounter Mr. Stalin, use of an example like (92a) is strange. The reason, I think, is that raising with this experiential verb is associated with the implicature that the entity denoted by the verb’s logical subject (here me) has had some direct experience of the entity (if any) denoted by the raised DP. On the contrary, (92b), which involves no raising and hence no implicature, is neutral and usable by anyone, regardless of his or her experience with Mr. Stalin. Since cases with seem are, though, the archetype of generally accepted raising structures, it would follow that the implicature linkage is not an indication of the nonexistence of raising, hence not an argument for a control analysis, for example. Taking that as the case, there is no reason to think that facts like (90) attack the view here that the on verb structures are instances of raising, not of control. Still, there is a mystery in (90b) beyond the mere facts of implicature linkage, namely, why the anomaly. This contrasts, for example, with seem structures: (93)

a. Stan seemed to Glenda to have defended Betty b. Betty seemed to Glenda to have been defended by Glen.

Here the b. example in no way seems anomalous. Not really understanding the nature of the implicatures linked to the raising cases, I can do no better than speculate that the difference lies in the nature of the linked assumptions. In the on verb cases, I suspect, as already speculated almost thirty years ago, that the assumptions involve reference to the will of the entity denoted by the raised DP, while this is not the case in the seem assumption. If so, then possibly what is wrong with (90b) is that the state


107

of affairs denoted by a passive is normally not attributable to the will of the entity denoted by the subject of the passive, even if there is such an entity and it has a will. In any case, I suggest that while the on verb cases do not provide a perfect model of the situation, Rosenbaum (1967) took to be a diagnostic for raising, neither do other cases of the sort that are almost uniformly taken to be raising cases. Hence I do not believe the implicature facts bear negatively on the conclusions of this chapter.

6. The structure of RCP cases The mere recognition of the existence of a raising analysis for cases like (94) leaves many questions about the relevant structures open: (94)

Winston counted/depended/relied on Isabelle to do the taxes.

In particular, despite what has been argued here, there is in the proper terms no reason to think that the raising involved in RCP actually involves anything that is accurately described as raising to the object position of a P. Although the claim just made might seem paradoxical, in relational terms, as already discussed in Postal (1986b) and Joseph (1990), it is not. In relational terms, raising itself means (i) that a constituent that heads an arc, say A, in a subordinate constituent (here, an infinitival clause) also heads at least one arc, say B, in the main clause, and that (ii) in a specifiable sense, arc B’s status depends on A. In the framework of Johnson and Postal (1980) and Postal (1990a, 1992, 1996), briefly sketched in chapter 1, section 9, which recognizes the primitive relations between arcs called Sponsor and Erase, these conditions can be taken to mean that A sponsors arc B, which erases arc A; technically then, B is a type I foreign successor of A. In such terms then, the reason that the raised DP ends up internal to a PP has to do with the relation that the raisee ends up bearing in the main clause. Simply put, that relation, call it R10, can be taken to be one that in English, unlike, for example, the core relations 1 (subject), 2 (direct object), and 3 (indirect object), requires the PP structure. There are then two possibilities. The more attractive, I believe, is to say that the raising with on verbs is simply raising to the same status as with more standard raising to object verbs like believe, that is, to direct object, 2. This means, in terms of the sketch of the previous paragraph, that B is a type I 2-arc foreign successor of A. One can then posit that on verbs obligatory require (at least certain) 2-arcs to have R10-arc type I local successors. It is possible and maximizes generality to assume that nonraising cases with the same verbs, like (95), then also involve R10-arc type I local successors of 2-arcs: (95)

Jenny counted/depended/relied on Lucy.

This means that one can take these verbs to occur with initial 2-arcs subject to the same requirement of demotion to R10 as holds for the raising cases. The difference is then that in (95) it is initial 2s that demote to R10, while in the cases discussed throughout this chapter it is 2s that result from raising (to object).

108


Viewed from this perspective, one can easily see that any analog of principles that ban RCP, principles argued here to be untenable just because of English cases like, for example, (94), would be quite strange. Such principles would merely stipulate that, for example, 2-arc foreign successors cannot have local successors with any R-sign that defines a relation that must determine a PP structure. Such a principle would be strange inter alia because the notion of ‘relation that must determine a PP structure’ does not pick out a universally specifiable set of relations. A particular relation may define a PP structure in one NL but not in another. The point then is that internal to the sort of arc-based ideas relevant to this section there is no independent motivation for any principle that accomplishes the work claimed to be done in LFG, HPSG, and GB frameworks by the various framework-specific assumptions that putatively block RCP. But without the needed stipulations, arc-based ideas, unlike these three views, arguably will not block the sort of raising that results in RCP. This is an obvious virtue if, as argued in earlier sections, RCP is instantiated in English. While the view that cases like (94) involve raising to 2 with subsequent demotion to R10 is attractive, it is not required. A distinct arc-based description would simply say that with on verbs raising is to R10; that is, infinitival complement 1-arcs have type I R10-arc foreign successors, which then determine the PP structure. Again, if this is not possible, in such terms it is only because of apparently fairly stipulative axioms that limit the class of foreign successors of, in particular, 1-arcs, axioms that might exclude R10-arc foreign successors. If such axioms could be justified, then the alternative just mentioned would be excluded and something like the raising to 2 with subsequent demotion to R10 might be imposed as the only available raising analysis.

7. Conclusion While the three most widely invoked syntactic frameworks have, as shown in earlier sections, all claimed as a consequence of fairly deeply rooted principles of those frameworks that RCP cannot be a feature of an NL, it has been argued in this chapter that such a view is falsified merely by an examination of certain facts of English. Such a conclusion then serves as a solid pillar for the fundamental claim of this work that enormous skepticism is justified with respect to current syntactic claims and conclusions.

3

A New Raising Mystery

1. Background Some quarter of a century ago a debate took place about the proper analyses to be assigned to various English clauses with nonfinite complements, including those like (1): (1)

a. b. c. d. e.

The doctor considered the condition to be untreatable. Most observers perceived them to be nervous. Isabelle proved Jerome to have vampirelike properties. Lydia wants Ken to succeed. The general wishes you to stand at attention.

At issue inter alia was the superficial clausal status of the highlighted DPs, hereafter the pivot DPs. Principal works involved include Bach (1977), Bresnan (1976), Chomsky (1973, 1981), Lightfoot (1976), and Postal (1974, 1977). Under the view defended in Postal (1974, 1977) and a bit in Postal and Pullum (1988), the pivot DPs were in one aspect of the structure of such sentences subjects of the complement clauses but in another aspect of that structure objects of the main clause. Under the view advocated in Chomsky (1973), the pivot DPs were exclusively subjects of the complement in every relevant structural aspect. Other positions are logically possible; for example, Pollard and Sag (1994: chapter 3) claim that the pivot DPs in (1) are exclusively main clause objects. It is unclear where these matters stand today. With the passage of time, the grammatical assumptions and theoretical commitments that underlay the debate of the 109

110


1970s have evolved; no one involved in those discussions holds today a theoretical position entirely like, perhaps even much like, that they held then. Moreover, new facts have of course come to light. It would thus be appropriate to reconsider the whole web of issues that surrounds cases like (1).1 But that task greatly exceeds the scope of this chapter. Rather, I focus here on a specific range of data that played no role in the earlier discussions but I believe bear significantly on the proper analysis of these constructions. Before getting to that, though, let me sketch informally my own current position on these matters. This is not so easy for the major reason that I think about such issues in terms of a relational or arc-based view of syntactic structure that is not widely known. Anyone interested should consult works like Johnson and Postal (1980), Postal (1990a, 1992, 1996), and chapter 1, section 9. It is not possible to discuss this conception in detail here. That is a major reason that what I will say will be quite informal. The original controversy took place internal to a certain number of assumptions that defined what was then called transformational grammar. Not accidentally, the subtitle of Postal (1974) included the phrase “One Rule of English Grammar,” a reference to a particular transformational operation whose existence was argued for. I have, however, long since rejected the idea that transformational rules, whether schematized as in Chomsky’s Move a formulations of recent decades or not, play any role in NL. More generally, I claim that there is no motivation for any generative apparatus at all. An NL grammar should, I believe, be regarded as an axiom system whose elements are statements to which truth values can be assigned, not as a sentence formation machine implementable as a computer program. I cannot defend this view here, but I hope to have at least made it clear that among the ideas I reject are those that both the opposing position and mine shared in the 1970s debate. See chapters 6 and 13 for further discussion of the issue of generative versus nongenerative grammars. Despite these differences from my 1974 framework, much of the substance of the earlier raising view of the relevant English clauses can be extracted from a transformational position and formulated in terms of a nongenerative conception of grammars and an arc-based conception of sentential structure. And if this is done, I believe the essence of the 1974 claims to be correct, even though some of the arguments for them were unsuccessful. The preservable substantive view is informally that stated at the beginning. The pivot DPs in cases like (1) are in one aspect of sentence structure subjects, in fact, typically what relational frameworks call final subjects, of the complement clauses, but also noninitial objects of some kind of the main clause. In arc terms, the pivot DPs head 1-arcs in the complement and noninitial 2-arcs in the main clause. Incidentally, one simple argument for a raising analysis that was not unsuccessful is that based on the particle-positioning facts in (2): (2)

a. Helen made there out to be seven gorillas in the clearing. b. Helen made it out to be seven miles to the next gas station. c. Helen figured it out to be impossible to square most circles.

The point was that the highlighted main verbs and particle out are lexically linked even though the latter follows the pivot DP. Under the position taken by Chomsky (1973, 1981) then, an element lexically linked to the main verb would have to be

A NEW RAISING MYSTERY

111

taken to appear inside the complement clause, while under a raising analysis it can be taken to be a normally positioned main clause particle. Remarkably, no defense of Chomsky’s position has, as far as I am aware, ever addressed this argument. By including cases like (1d and e) under the raising to object umbrella, I enter into especially controversial territory, for a number of writers, most notably Bach (1977) and Lightfoot (1976) in their reviews of Postal (1974), claimed that whatever the status of the raising claim for cases like (1a–c), which involve what I called B-verbs, that for cases like (1d and e), which involve what I called W-verbs, is much worse. Both invoked an earlier analysis from Bresnan (1972), which appealed to distinctions in complementizer status for the two cases. This line of thinking receives a modern interpretation in Lasnik and Saito’s (1991) work, which recognizes some kind of raising for the B-verb case while still denying it for the W-verb structures. The view that the pivot DP is a main clause constituent in both B-verb and W-verb cases, contra the view of Chomsky (1973), is supportable, though, by the grammaticality of cases like (3): (3)

a. Herbert proved Henry without any difficulty to be a spy. b. Herbert desires/wants/wishes Henrietta with all his heart to recover.

Under any straightforward implementation of Chomsky’s view, the highlighted main clause modifying adverbials are inside a subordinate clause, an IP in some instantiations, which is of course otherwise unknown. As a last background remark, I should state that my current view actually recognizes a scope for the raising of subjects into nonsubject status in English even broader than was central to the earlier discussion. For, as discussed in detail in chapter 2, I take all of the prepositional phrase cases in (4) to instantiate such raising: (4)

a. b. c. d.

We can’t count on there to be enough beer to keep all the students happy. One cannot depend on the lovebug to bite him just when it would be convenient for us. You can rely on all hell to break loose when the regional director shows up. Stan arranged for there, despite my misgivings, to be a meeting with the students.

Such cases were barely touched on in the 70s. But as was discussed in chapter 2, not only Chomsky’s GB framework but also the HPSG framework of Pollard and Sag and Bresnan’s LFG framework have all been explicitly constructed so as to bar a raising analysis of cases like (4); apparently the former two frameworks would impose instead a control analysis, while the latter appeals to a variant of ‘reanalysis’ that is not viable. The expletive pivot DPs in (4a and d) and the idiom chunk pivot DPs in (4b and c) already cast considerable doubt on a control analysis, and as shown in chapter 2, there is much other evidence as well against a control view of such cases.

2. Some puzzling facts The central topic of this chapter is introduced by data like (5): (5)

a. (I am sure that) The woman who is favored to win screamed. b. (I am sure that) The woman who is favored to win is a doctor. c. (I am sure that) The woman who is favored to win is you.

112


English clauses like (5a–c) seem initially parallel; each apparently involves a different type of predicate phrase that occurs with an instance of the same subject. But these apparently identical subjects fail to behave identically in a variety of constructions, as shown in (6)–(10); here and throughout, indexed subscripted letters t, pg are used to indicate extraction gaps/parasitic gaps, with no theoretical implications about the existence of traces, and so on. (A) Topicalization (6)

a. [The woman who is favored to win]1 I am sure t1 screamed. b. [The woman who is favored to win]1 I am sure t1 is a doctor. c. *[The woman who is favored to win]1 I am sure t1 is you.

(B) Clefting (7)

a. It is the woman who is favored to win who1 I am sure t1 screamed. b. It is the woman who is favored to win who1 I am sure t1 is a doctor. c. *It is the woman who is favored to win who1 I am sure t1 is you.

(C) Object Raising (8)

a. [The woman who is favored to win]1 is hard to believe t1 to have screamed. b. [The woman who is favored to win]1 is hard to believe t1 to be a doctor. c. *[The woman who is favored to win]1 is hard to believe t1 to be you.

(D) Object Deletion (9)

a. [The woman who is favored to win]1 is too old for us to believe t1 to have screamed. b. [The woman who is favored to win]1 is too old for us to believe t1 to be a doctor. c. *[The woman who is favored to win]1 is too old for us to believe t1 to be you.

(E) Parasitic Gaps (10)

a. the guy who1 they will arrest t1 after proving pg1 to have screamed b. the guy who1 they will arrest t1 after proving pg1 to be a doctor c. *the guy who1 they will arrest t1 after proving pg1 to be you

While initially anomalous-seeming, the differences in (6)–(10) can apparently be reduced to regularities under well-motivated assumptions, given the facts in (11), which show up when one replaces the subjects of the relevant clauses in (5) with weak definite pronouns: (11)

a. (He noticed [the woman who is favored to win]1 because) she1 screamed. b. (He noticed [the woman who is favored to win]1 because) she1 is a doctor. c. *(He noticed [the woman who is favored to win]1 because) she1 is you.

These data show that the subject position in (5c) but not those in (5a and b) is an antipronominal context in the terminology of Postal (1993a, 1993b, 1994a, 1994b,


113

1998, 2001a, 2001b), a position from which weak definite pronouns are, for whatever reason, barred. This can explain the contrasts in (6) and (7) under the assumption argued in Postal (1994a, 1998) that both DP topicalization and DP clefting in English are what were called B-extractions. These obligatorily link to invisible resumptive pronouns in their extraction sites. For if, as also argued in some of those publications, resumptive pronouns are weak definite pronouns, it follows that B-extraction sites and antipronominal contexts should in general fail to intersect. The works just cited argue that this is the case, but see Levine (2001) for a skeptical rejection of these ideas. I believe that what follows not only appeals to the notion of B-extractions but also in fact supports their recognition. Six examples that involve antipronominal contexts distinct from that in (5c) are given in (12)–(17): (12)

a. b. c. d. e.

Herman was speaking French/it. Herman was speaking in French/*it. [Which language]1 was Herman speaking (in) t1? French1, Herman was definitely speaking t1. *French1, Herman was definitely speaking in t1.

(13)

a. b. c. d.

Ethel was inside of the sphere/it. Ethel was inside the sphere/*it. [Which sphere]1 was Ethel inside (of) t1? [That sphere]1, Ethel was inside *(of) t1.

(14)

a. Marjorie quit that police unit/*it/*them. b. [Which police unit]1 did Marjorie quit t1? c. *[That police unit]1, Marjorie quit t1 last week.

(15)

a. She gave those walls a coat of paint because he would not give them a coat of paint/*it. b. [Which coat of paint]1 did she give those walls t1? (answer: the second). c. *[That coat of paint]1, I gave those walls t1.

(16)

a. There were particles of plutonium/*them in the pudding. b. [Which kinds of particles]1 were there t1 in the pudding? c. *[That kind of particles]1, there were t1 in the pudding.

(17)

a. She was unable to tell (= ‘determine’) the distance to the sound/*it. b. [Which distance]1 was she unable to tell t1? c. *[The distance to the sound]1, she was unable to tell t1.

The contrasts in (12)–(17) receive a parallel explanation from the same underlying pronoun contrasts on the assumption that DP topicalization is a B-extraction. Moreover, if it is accepted, as claimed in, for example, Cinque (1990) and Postal (1994a, 1994b, 1998, 2001a, 2001b), that the gaps in the object raising, object deletion, and parasitic gap constructions also are invisible weak definite pronouns, the

114


contrasts in (18)–(23) fall out, as these constructions, too, cannot have as their gaps the antipronominal contexts in (12)–(17): (18)

a. [That language]1 is difficult to speak (*in) t1. b. *[That language]1 is too complex for me to speak (*in) t1. c. [Which language]1 did he criticize t1 while speaking (*in) pg1?

(19)

a. [That sphere]1 is impossible to get inside *(of) t1. b. [That sphere]1 is too small to get inside *(of) t1. c. [Which sphere]1 did Melissa criticize t1 after finding herself inside *(of) pg1?

(20)

a. *[That police unit]1 was difficult for Marjorie to quit t1. b. *[That police unit]1 is too well paid for Marjorie to quit t1. c. *[Which police unit]1 did Helga join t1 immediately after Marjorie quit pg1?

(21)

a. *[The second coat of paint]1 was difficult for her to give the walls t1. b. *[The second coat of paint]1 will be too thick for a child to give those walls t1. c. *[Which coat of paint]1 did Marsha criticize t1 after Sally gave those walls pg1?

(22)

a. *[Those particles]1 were impossible for there to be t1 in the pudding. b. *[Those particles]1 are too rare for there to be t1 in the pudding. c. *[Which (type of) particles]1 did there being pg1 in the pudding lead the FDA to ban t1?

(23)

a. *The distance to the sound was impossible for her tell. b. *The distance to the sound was too far for her to be able to tell. c. *[What distance]1 did her being unable to tell pg1 lead Jim to try to estimate t1?

The background facts so far might not in themselves merit much further discussion. What does, though, is data not, I believe, previously noticed that relates to previous claims in Postal (1994b) formulable informally as in (24) and (25): (24)

Right Node Raising DP gaps, in contrast to those that involve, for example, topicalization or clefting DP gaps, are not sensitive to antipronominal contexts.

(25) Complex DP Shift gaps are not sensitive to antipronominal contexts.

These claims, linked to the view that, unlike, for example, topicalization, neither Right Node Raising nor Complex DP Shift requires resumptive pronouns, indicate that nothing should preclude intersection of the classes of Right Node Raising and Complex DP Shift gaps with antipronominal contexts. Claim (24) can be supported by data like (26), which involves antipronominal contexts already mentioned: (26)

a. Ted may have spoken (in) t1 and Sandra certainly did speak (in) t1 [that very obscure Oriental language]1. b. Ted may have gotten inside (of) t1 and Sandra certainly did get inside (of) t1 [that very peculiar-looking sphere]1.


115

c. Marjorie certainly quit t1 last week and Jane will probably quit t1 this week [that very prestigious police unit]1. d. Lester may have given those walls t1 and he certainly gave these walls t1 [a second coat of paint]1. e. There may be t1 in the first sample and there certainly are t1 in the second sample [the sort of particles at issue]1. f. Ted may have been unwilling to say she was unable to tell t1, but she was unable to tell [the distance to the overheard sound]1.

Claim (25) cannot be supported from all of the same antipronominal contexts given the fact first noted in Ross (1967/1986) that Complex DP Shift gaps cannot be complement DPs of prepositional phrases. But (25) is still easily supportable, as in (27): (27)

a. b. c. d. e. f. g. h.

Ted dyed his eyebrows green/that color/*it yesterday. Ted dyed his eyebrows t1 yesterday [an awful shade of bright yellow]1. Ted gave that idea a lot of consideration/*it yesterday. Ted gave that idea t1 yesterday [more consideration than it deserved]1. Marjorie quit t1 last week [that very prestigious police unit]1. Molly will give that wall t1 tomorrow [the second coat of paint it needs]1. There were t1 in the pudding [exactly the sort of particles we had feared]1. Ted was unable to tell t1 with any exactitude [how far the alien planet was from our solar system]1.

Given this background, the contrasts in (28) and (29) are initially rather bewildering: (28)

a. Ted may believe t1 to have screamed and Archie does believe t1 to have screamed [the woman who is favored to win]1. b. Ted may believe t1 to be a doctor and Archie certainly does believe t1 to be a doctor [the woman who is favored to win]1. c. *Ted may believe t1 to be you and Archie certainly does believe t1 to be you [the woman who is favored to win]1.

(29)

a. I believe t1 to have screamed [the woman who is favored to win]1. b. I believe t1 to be a doctor [the woman who is favored to win]1. c. *I believe t1 to be you [the woman who is favored to win]1.

For the good and bad cases of this Right Node Raising and Complex DP Shift data match exactly the pattern of ordinary and antipronominal contexts in (5). That is, the ungrammatical cases of (28) and (29) would apparently be what one would expect if Right Node Raising and Complex DP Shift were somehow sensitive to antipronominal contexts. But this is just what (24) and (25) deny, and, as I have illustrated, with real factual motivation. This seemingly points to the ugly conclusion that (28c) and (29c) are bad for (unknown) reasons that have nothing to do with (5) and hence nothing to do with the parallel instances of ill-formedness in the topicalization, clefting, object

116


raising, object deletion, and parasitic gap paradigms. Such an ‘accidental similarity’ conclusion is obviously thoroughly implausible, even for a single antipronominal context case like that in (5c). And the dubious status of an ‘accidental similarity’ view worsens when it is observed that the same state of affairs exists for other subject antipronominal context cases, like those in (30): (30)

a. Something is the matter with his fuel pump (*but it is not the matter with mine). b. Lots of things are wrong with my liver (*but fortunately they are not wrong with your liver).

As expected, these antipronominal sites are incompatible with topicalization, object raising, and so forth; see (31): (31)

a. *[Something really terrible]1 they claimed t1 was the matter with the fuel pump. b. *[Something like that]1 was impossible to believe t1 to be the matter with the fuel pump. c. *It was that which1 they claimed t1 was wrong with her liver. d. *That1 is too rare to believe t1 to be wrong with her liver.

And just like the antipronominal context in (5c), those in (30) also correlate with bad Right Node Raising and Complex DP Shift gaps in cases parallel to (28) and (29), that is, in putative raising to object structures like those of (32): (32)

a. *Ted may have claimed t1 to be the matter with the fuel pump and Joe did claim t1 to be the matter with the fuel pump [something really terrible]1. b. *Ted claimed t1 to be the matter with the fuel pump [something probably irreparable]1. c. *The doctor may have believed t1 to be wrong with her liver and the nurse did believe t1 to be wrong with her liver [something that required immediate surgery]1. d. *The doctor believed t1 to be wrong with her liver [something really terrible]1.

One can summarize the discussion so far as follows. Where it was previously claimed that neither Right Node Raising nor Complex DP Shift gaps are incompatible with antipronominal contexts, conclusions supported consistently by data like (26) and (27), at least one class of subject antipronominal contexts does seem to induce Right Node Raising and Complex DP Shift violations under specific conditions, namely, where these define instances of the raising to object construction. So (33) seems to be the case: (33)

If a is a subject antipronominal context and X is a putative raising to object structure whose complement subject occurs in a, then the derived object position in X is not compatible with Right Node Raising and Complex DP Shift gaps.

Assuming that (33) is essentially true, a conclusion for which further evidence will be provided, it can hardly be an accident. The incompatibility of certain raised subjects with Right Node Raising and Complex DP Shift in these particular circum-


117

stances must relate to the existence of the relevant subject antipronominal contexts This creates an apparent paradox, given (24) and (25), which state that Right Node Raising and Complex DP Shift in general are indifferent to antipronominal contexts and which are supported by considerable data independent of the raising to object constructions at issue.

3. An initial approach to the paradox One would like to maintain (24) and (25) without any ad hoc exceptions for the raising to object construction or any other and yet still reduce generalization (33) to the fact that the raised phrases are subjects that occur in antipronminal contexts. Securing this result involves initially four descriptive elements, informally describable as follows: Element one is to continue to assume, as in previous work of mine, that topicalization is, but Right Node Raising and Complex DP Shift are not, in general sensitive to antipronominal contexts because the former must, but the latter need not, link to (evidently invisible) resumptive pronouns in their extraction sites. This means that topicalization is a B-extraction but that Right Node Raising and Complex DP Shift contrast in not determining resumptive pronouns. Element two is to assume that the constructions at issue do involve raising to object. Focus for concreteness on, for example, (34d): (34)

a. b. c. d.

Ted claimed it to be imminent. Ted claimed t1 to be imminent [something probably irreparable]1. Ted claimed something probably irreparable to be the matter with the fuel pump. *Ted claimed t1 to be the matter with the fuel pump [something probably irreparable]1.

Given element two, both (34b and d) will involve raising of the lower subject out of the lower clause and into the main clause. Element three consists of informal assumption (35): (35)

In the cases where Right Node Raising and Complex DP Shift are bad, the instance of raising to object involves its own proper (invisible) resumptive pronoun.

Claim (35) means that in (34d) the raising of something probably irreparable out of the complement requires the presence of a resumptive pronoun as subsequent complement subject, although this is not the case in (34b), where the raising is from a context that is not antipronominal. So, under current assumptions, the pronouns that end up violating the antipronominal context condition on the subject of, for example, be the matter with are resumptive pronouns linked to raising to object, not resumptive pronouns linked to Right Node Raising or Complex DP Shift. This means inter alia that in a fuller schematic structure of (34d) there is a resumptive pronoun in the complement but none in the main clause: (36)

*Ted claimed t1 – [something probably irreparable]1.

118


The fact that the resumptive pronoun is an element of the infinitival complement is what guarantees consistency with principle (25). For the Complex DP Shift gap is in the main clause, while the antipronominal context is in the complement. Significantly, claim (35) cannot be interpreted to mean that the presence of raising-linked invisible resumptive pronouns is uniformly associated with raising to object, even for the very main verbs like believe at issue and even for the very antipronominal subject complement types under consideration. The impossibility of such a general conclusion is shown as follows: A variety of raising to object structures whose defining elements raise out of subject antipronominal contexts are perfectly grammatical when the raised form is not extracted under Right Node Raising or Complex DP Shift. The latter condition is met if the raised phrase is either left extracted or not extracted at all. So all of (37) are fine, although the lower subject is, as already seen, in each case an antipronominal context: (37)

a. b. c. d. e. f. g.

Ernestine believes the person favored to win to be you. Ernestine believes something grave to be the matter with the fuel pump. Ernestine believes something probably fatal to be wrong with her liver. Who1 did Ernestine believe t1 to be you? [Nothing of the sort]1 did Ernestine believe t1 to be the matter with the fuel pump. [No matter what]1 Ernestine believed t1 to be wrong with her liver, . . . Something grave was believed to be the matter with the fuel pump.

To maintain consistency with this antipronominal character, cases such as (37), like (34b) and significantly unlike (34d), must not involve complement subject resumptive pronouns. Therefore, a required element four is as in (38): (38)

Although (a) raising to object can in general link to a complement subject resumptive pronoun or not, (b) a linked subject resumptive pronoun in the complement is obligatory if the object formed by raising serves as the extractee for Right Node Raising or Complex DP Shift.

That is, an absolutely essential feature of this four-element proposal about the raising to object cases is that association of a resumptive pronoun with an instance of English raising to object is conditional on whether or not the raisee is an extractee in the higher constituent and, moreover, an extractee of the Right Node Raising or Complex DP Shift types. While there are syntactic views that could not incorporate such a ‘nonderivational’, ‘noncyclic’ condition, happily a relational framework of the sort described in Johson and Postal (1980), Postal (1990a, 1992, 1996), and chapter 1, section 9, faces no problems with such an account, which would be quite straightforward.2 I will argue, though, that what is involved is more general and is not limited to raising to object cases or to those that involve Complex DP Shift or Right Node Raising.

4. Expanding the database The factual grounding of conclusion (38) is stronger than indicated so far. Further support will derive from any other more or less lexical type of subject antipronominal


119

context. For (38) predicts that any such subject that enters into the raising to object construction will fail to permit Right Node Raising or Complex DP Shift of the raised subject. Christopher Potts has pointed out to me several further cases that include, first, the expression in (39a): (39)

a. b. c. d.

Nothing/Very little/Little of value/*It came of that proposal. He believes little of value to have come of that proposal. *He believes t1 to have come of that proposal [little of value]1. *Frank may believe t1 to have come of that proposal and Gloria does believe t1 to have come of that proposal [little of value]1.

Just as principle (38) predicts, both types of right extraction structures are ill formed. Second, Potts observes the expression in (40a): (40)

(Scientist notices his lab rat has gained incredible strength after its injection): a. What! Something strange is going on here! But that/*it was not going on before the injections. b. *Dr. Frankenstein believes t1 to be going on in his lab [something quite strange]1. c. *Dr Frankenstein may believe t1 to be going on in his lab and he certainly believes t1 to be going on in your lab [something quite strange]1.

Again the otherwise anomalous gaps in Right Node Raising and Complex DP Shift paradigms are as predicted by (38). Third, Potts notes that the sense of the verb eat roughly equivalent to that of bother found in (41a) also has an antipronominal context subject, as supported by the data in (41b): (41)

a. What’s eating/bothering Gilbert Grape? b. Something is eating/bothering you, Gilbert; I can tell. (But whatever it is, it is not *eating/bothering your sister.) c. Donna believes t1 to be *eating/bothering Gilbert [something incomprehensible to her]1. d. Donna may believe t1 to be *eating/bothering Gilbert and she certainly believes t1 to be *eating/bothering Neil [something incomprehensible]1.

And once more the Right Node Raising and Complex DP Shift paradigm holes predicted by (38) are found.

5. Links to raising to subject structures So far I have supported descriptive principle (35) with six instances of antipronominal subject contexts. These data, which involve what I consider raising to object structures, reveal certain striking parallels to relatively uncontroversial raising to subject structures, for example, with the adjectival raising to subject trigger likely. Consider, for example, (5’) parallel to (5) and (28’) parallel to (28) and (29):

120 (5')

(28')


a. The woman favored to win is likely to scream. b. The woman favored to win is likely to be a doctor. c. The woman favored to win is likely to be you. a. How likely to scream is the woman favored to win? b. How likely to be a doctor is the woman favored to win? c. *How likely to be you is the woman favored to win?

The antipronominal context, which in (28) and (29) precludes the raised subject in a raising to object structure from being an extractee for Right Node Raising or Complex DP Shift, in these raising to subject cases is incompatible with the fronting of the adjectival phrase that includes the raising trigger. To see that this parallelism is nonfortuitious, observe that a parallel holds for other antipronominal contexts cited for the object case, as shown in (32'), (39'), (40'), and (41'): (32')

a. b. c. d. e. f.

Something serious was likely to be the matter with the fuel pump. *How likely t1 to be the matter with the fuel pump was [something serious]1? *Likely to be the mattter with the fuel pump though [something serious]1 was, . . . [Something really grave] is likely t1 to be wrong with her liver. *How likely t1 to be wrong with her liver is [something really grave]1? *Likely to be wrong with her liver though [something really grave]1 was, . . .

(39')

a. Nothing (much)/Very little/Little of value is likely to come of that proposal. b. They claimed little of value was likely to come of that proposal (*and likely t1 to come of that proposal [little of value was]1. c. *Likely t1 to come of that proposal though [little of value]1 was, . . .

(40')

a. Something unexpected is likely to be going on in that lab. b. *How likely t1 to be going on in that lab is [something unexpected]1?

(41')

a. Something seemingly trivial is likely to be eating/bothering Gilbert. b. How likely t1 to be *eating/bothering Gilbert is [something seemingly trivial]1? c. Likely t1 to be *eating/bothering Gilbert though [something seemingly trivial]1 was, . . .

The good variants of (41'b and c) already show that there is no general ban on combining the sort of frontings at issue with raising to subject structures. This is evidenced further in (42) and (43): (42)

a. b. c. d. e.

How likely t1 to show up late are [some of your students]1? How likely t1 to puzzle a child is [something that strange]1? How likely t1 to raise eyebrows is [something so risque]1? Likely t1 to show up late though [some of your students]1 are, . . . Likely t1 to puzzle that child though [something that strange]1 is, . . .

(43)

a. They said that something serious usually tends to be the matter with the carburetor/ to happen at the wrong time.


b. c. d. e. f.

121

*and tend t1 to be the matter with the carburetor, [something serious]1 usually does. and tend t1 to happen at the wrong time [something serious]1 usually does. Something like that is likely to be wrong with her liver/harmful to her liver. How likely t1 to be harmful to her liver is [something like that]1? *How likely t1 to be wrong with her liver is [something like that]1?

The raising to subject data apparently ground a descriptive generalization like (44): (44)

If a is a subject antipronominal context and X is a putative raising to subject structure whose complement subject occurs in a, then the adjectival/verbal phrase that contains the raising trigger cannot be left extracted.

6. Generalizing over raising to object and raising to subject structures Given that it has been found that constraints similar to those for raising to object cases hold for raising to subject ones, questions like those in (45) arise: (45)

a. Why should the subject antipronominal contexts that were looked at in connection with putative raisings to object apparently interact with the adjectival and verbal phrase frontings to seemingly block certain raising to subject cases? b. What structural property is shared by (i) Right Node Raising and Complex DP Shift interactions with raising to object and (ii) Adjectival Phrase fronting/raising to subject interactions such that both (38) and (44) hold?

Under the assumption that the sets of structures I have looked at, those with putative raising to object main verbs and those with raising to subject triggers like likely, both involve raisings, the following emerges: When a raised subject in the former case is the extractee for Complex DP Shift or Right Node Raising, the raised DP then linearly follows its own extraction site, although it does arguably c-command it just as in a non-Complex DP Shift or Right Node Raising case. Moreover, when the whole adjectival phrase that contains a raising to subject trigger is left extracted under questioning, though extraction, and so forth, the raised DP both follows its own extraction site and fails to c-command it. I conclude that at least one of these two partially correlated properties is crucial in predicting the bad results. But there is a deeper issue. Namely, since the linear precedence facts are common to all the cases and yet so far ungrammaticality has only been attested in those where the raising source positions are antipronominal contexts, a key puzzle is the linkage between antipronominal contexts and the constructions that influence word order. The crucial issue, I believe, is to understand how and why antipronominal contexts induce ungrammaticality in just a proper subset of the variety of cases I have documented. I would like to propose that the linkage is essentially due to a generalization of the point inherent in the four-element proposal made earlier about just the raising to

122


object case. Namely, the structural conditions at issue, which involve Right Node Raising and Complex DP Shift in the raising to object case and left extractions of adjectival/verbal phrases in the raising to subject case, interact with antipronominal contexts because under the very structural conditions at issue it is required that either type of raising leave a resumptive pronoun in the complement subject position. These are just the positions where antipronominal contexts have been shown to be associated with ungrammaticality. In short, just as I claimed earlier that, for example, structures like (46a and b) have partially different complement structures in which (46a) need not but (46b) must involve a resumptive pronoun, just so one can make a parallel claim for (47): (46)

a. They found something quite trivial to be bothering/eating Gilbert. b. They found t1 [something quite trivial]1.

(47)

a. Something quite trivial is likely to be bothering/eating Gilbert. b. How likely is [something quite trivial]1.

The posited resumptive pronouns provide the link between the reorderings and the restriction of ungrammaticality to just those cases where the raising source position is antipronominal.3 To develop the resumptive pronoun claim, it is of course necessary to give an account of the conditions under which raisings must link to resumptive pronouns. Since for reasons touched on at the beginning I am not proposing a precise framework for this description, I cannot approach this issue in a truly serious way. Still I suggest something like (48) is probably at work: (48)

The Raising Resumptive Pronoun Linkage Condition a. Let the notion raising be restricted for this discussion to the case where some DPx constituent of a complement clause C1 also appears as a noninitial subject, object, or oblique constituent of a constituent K that contains C1. Then: b. If DPx raises out of C1 into K from position P, a resumptive pronoun must appear in P unless all of (i)–(iv) hold: (i) DPx is the last nonresumptive subject of C1. (ii) C1 is nonfinite. (iii) If C1 is a constituent of a clause C2, distinct from K, K is a constituent of C2 (that is, C1 is the highest clausal constituent of K). (iv) The surface realization of DPx both c-commands and linearly precedes a surface realization of the (raising) remnant of C1.

I must leave it open here whether (48) should be interpreted as a universal or merely a principle of English.4 But the idea is that in ordinary raising to subject cases like that with English likely, all of (48bi–iv) will hold but that a legitimate NL raising construction can fail to satisfy any or all of them provided that a resumptive pronoun, visible or not, appears in the raising origin site. Put differently, I am suggesting that (48bi, ii, iii, and iv) are the conditions required for raising that does not involve a resumptive pronoun.


123

To briefly go over the intended claims, (i) essentially picks out the case where the raisee is a subject; (ii) is self-explanatory, while (iii) specifies that the highest clause from which raising occurs is is an immediate constituent of the clause of which the raised element is a noninitial constituent. Condition (iv) is the one crucial for this discussion. It requires that any raisee that does not link to a resumptive pronoun bear two fixed structural relations to the remnant clause out of which it raises and, moreover, requires these relations to hold at the surface level. Although (48biv) mentions c-command, there is in fact nothing in the English data gone over that really motivates that. For in every bad case of relevance that has been cited, the raisee ends up linearly following the raising remnant. Moreover, arguably, in the case of, for example, raising to object where the raisee is the target of Complex DP Shift or Right Node Raising, the raisee continues to c-command the site of origin for raising. My guess, though, is that for (48biv) to have any cross-linguistic plausibility, something nonlinear would be required. What (51biv) says is that any grammatical phenomenon that brings about a raising remnant clause not being both c-commanded and linearly preceded by the raisee requires that the instance of raising involve a resumptive pronoun. This will inter alia generate ungrammaticalities in cases where the raising site is an antipronominal context. There are relevant phenomena of a sort rather different from those so far illustrated that have effects like those referenced by (48biv), for example, the clausal and verbal phrase ellipsis phenomena in (49): (49)

They said Carol was sick, a. as she is/b. which she is/c. and so she is/d. and that she is/ e. and she is.

If (48) is on the right track, all of these should yield ungrammaticalities when combined with raisings. That is, one should for instance find contrasts in cases like (50): (50)

They said something was bothering/eating Gilbert, as something was/b. which something was/c. and so something was/d. and that something was/e. and something was.

And cases like (51) should just be outrightly bad: (51)

They said something of the sort was wrong with the transmission a. as something of the sort was. b. which something of the sort was. c. and so something of the sort was. d. and that something of the sort was. e. and something of the sort was.

These are quite subtle data, though, and I leave it to readers to make their own judgments. While a condition like (48) is motivated by data of the sort I have gone over that involve raising to subject and raising to object in infinitival cases, it has a much broader scope. Limiting attention only to English, it first of all naturally allows a raising analysis of the construction in (52):

124 (52)


Mike looks/seems/sounds like/as if he is a werewolf.

Since the highest clause out of which raising would occur in (52) is finite, (48) forces a resumptive pronoun, arguably represented by the highlighted form. A raising analysis of this construction is moreover supportable via traditional expletive and idiom chunk arguments, as first noted in Rogers (1973) and illustrated in (53) and (54): (53)

a. There look/*looks like/as if there/*it/*them/*her are/*is going to be problems with the dean. b. There *look/looks like/as if there *are/is going to be a problem with the dean.

(54)

a. b. c. d. e. f. g.

The chickens look like they have finally come home to roost. Martha’s ship looks like it has come in. The wolf looks like it is at the door. The lovebug sounds like it has bitten Mary again. The shit seems like it is about to hit the fan. The ball sounds like it is in your court. Fortune seems like it has finally smiled on Myriam.

I would take the complement occurrences of there in (53) to be resumptive pronouns, the rule being evidently that raised there determines a resumptive of the same shape and that a resumptive pronoun agrees with the element it ‘resumes’ in person, number, and gender. Perhaps the same shape condition can be subsumed under the rule operative for tags like that of (55): (55)

a. There are gorillas in the field, aren’t there/*it/*they? b. Into the bar there strode a mean-looking dude, didn’t there/*it/*they?

It is rather remarkable that one has seen decades of theorizing about limitations on raising with postulation of principles that would not allow a raising analysis of this construction, despite the fact that it has been known for years.5 Moreover, views that would not allow a raising analysis seem to have offered no alternative. Further, condition (48) is consistent with the fact that this construction seemingly has instances in which nonsubjects raise, although these are restricted, in ways I do not understand; see (56): (56)

a. Melissa sounds like Bob has been hassling her again. b. Melissa’s arm looks like the dog has been biting it again. c. Melissa looks like Bob has been kicking her in the arm again. d. ?Melissa sounds like people want to force her to resign. e. Melissa sounds like Bob believes her to have a chance of winning. f. *There sounds like Bob believes there to be no chance of her winning. g. *The wolf sounds like Bob believes it to be at the door.

Principle (48) can also be taken as a key component in an understanding of the object-raising construction, represented by the once famous sentence in (57):


(57)

125

John is easy to please.

If, as in early transformational accounts, this is analyzed in terms of raising, since the raisees are in general nonsubjects, (48) forces a resumptive pronoun. This predicts, correctly I believe, that the gaps for this construction are incompatible with antipronominal contexts, already partly supported earlier. One thing that has been claimed to argue against a raising analysis of this construction is that it does not allow the raisee to be an idiom chunk, in comparison, for example, to raising with seem; see (58): (58)

a. Chomsky (1981: 309) “We therefore expect it to be resistant to idiom chunks and other non-arguments . . .” b. Chomsky (1981: 309) *Good care is hard to take of the orphans. c. Chomsky (1981: 309) *Too much is hard to make of that suggestion.

But while true that such raising is far more restricted, that (58b and c) are bad, and that the construction absolutely bars raising of expletives, as Chomsky noted, there are nonetheless numerous acceptable instances of idiom chunk object raising, as illustrated in (59): (59)

a. b. c. d. e. f. g. h.

The baby is easy to throw out with the bathwater. The ice is difficult to break at faculty parties. The cat will be easy to prove to be out of the bag. The jury is easy to show to still be out on that proposal. The bottom is now easy to imagine falling out of the cocoa market. Thatcher’s shoes are impossible to imagine anyone like you filling. ?Stan’s ass is not hard to anticipate being in a sling again. The rug is impossible for me to imagine being pulled out from under a guy like that. i. Jacobson (1992b: 271) Careful attention was very hard to pay to that boring lecture. j. Jacobson (1992b: 271) ?The cat would be quite easy to let out of the bag. k. McCawley (1998: 115) *The cat was easy to let out of the bag. l. McCawley (1998: 107) John’s leg is easy to pull. m. Nunberg, Sag, and Wasow (1994: 517) The law can be hard to lay down.

Two further notes about object raising: While in American English it is, I believe, impossible to object raise a subject that has not independently raised to object, this is, I have learned, arguably not entirely so in British English. I heard an example essentially like (60) in a televised documentary with a British narrator: (60)

That attitude made [such a tragedy]1 all too easy t1 to happen.

I have checked this with two British speakers, both of whom seem to accept it, whereas for me, and I suspect most Americans, it is entirely impossible. One notes of course

126


that the subject in question is unaccusative, and brief investigation suggests that this is required. Even those who accept (63) seem to firmly bar (61): (61)

*That attitude made [such a tragedy]1 all too easy t1 to devastate Rodney.

I conclude then that both the finite construction of (55) and the object-raising construction are bona fide raising cases required by (48) to involve resumptive pronouns. Of course, even if true, this still leaves open key issues. An obvious one is the visibility of the resumptives in the former and their absence in all the other cases I have dealt with. This suggests that at least English has a principle like (62), where the notation refers back to (48): (62)

Raising-Linked Resumptive Pronoun Visibility A raising-linked resumptive pronoun is visible if and only if C1 is finite.

Note that quite properly, principle (62) requires an invisible resumptive pronoun in cases where object raising raises a phrase from a finite clause that is itself embedded in a nonfinite one, as in (63) for that subset of speakers, like me, who accept such: (63)

Michelle1 will not be that easy [C1 to inform them [C2 that you plan to dismiss (ok RP1 = Ø/*RP1 = her)]].

7. Earlier data There are two further issues I want to raise, one that involves a proposal by Lasnik and Saito and another that involves related but more general issues that concern idiomatic DPs. First, Lasnik and Saito (1992: 140–142) cite observations attributed to my colleague Mark Baltin that (64a) and (65a) are bad despite (64b) and (65b). Both contrast with (66a). (64)

a. *How likely t1 to be a riot is there1? b. There is likely to be a riot.

(65)

a. *How likely t1 to be taken of John is [advantage]1? b. Advantage is likely to be taken of John.

(66)

a. How likely t1 to win is [John]1. b. John is likely to win.

Lasnik and Saito proposed to account for (64a), (65a) and related data via a principle to the effect that a trace must be bound (hence c-commanded) by its antecedent, at every level; see (67): (67)

Lasnik and Saito (1992: 90) The Generalized Proper Binding Condition Traces must be bound throughout a derivation.


127

At first glance, there is a remarkable partial parallelism between the cases in (64) and (65) and the raising to subject cases I cited earlier, for example, (32'c), (39'b, c), (40'b) and (41'b). All involve ungrammaticality only when a raising to subject remnant is itself left extracted. The earlier cases involved raising of ordinary DPs from antipronominal contexts. The ones Lasnik and Saito deal with involve expletive and idiomatic DPs. One cannot help but inquire into whether the phenomena dealt with here and those for which Lasnik and Saito invoke (67) are the same. I find that there are solid grounds for thinking they are the same but also problematic issues that I do not currently know how to resolve and that could undermine this conclusion. I will briefly consider both the positive and negative aspects. Consider first the expletive cases revealed by the there data Baltin noted. Here arguably there does not occur in antipronominal contexts. Rather there is reason to consider there itself a weak definite pronoun, as argued by (53a, b) earlier. This might make it appear that there is no way to directly reduce the facts to principle (48). But this is not necessarily so. It is evident that, as required by principle (62), in all the infinitival cases when (48) requires a resumptive pronoun the latter must be invisible. Suppose as I have speculated elsewhere (see Postal, 1993a: 752–753; 1994b: 93–96; 2001a: 237–238) about other types of invisible resumptives, this is due to a type of control phenomenon. And suppose principle (68a) holds for everyone while some people have the still stronger (68b): (68)

a. The expletive there DP cannot control a resumptive pronoun. b. The expletive there DP cannot control any pronoun.

Principle (68b) is motivated by facts in standard adjunct control environments like those of (69): (69)

a. * (There) being a gorilla in a living room proves there to be a gorilla in the bedroom. b. */okThere can be war in the north without (there) being war in the south.

Everyone seems to reject control in (69a). But some speakers, including me, accept control in (69b), arguing that (68b) is too general. But even the weaker (68a) would combine with (48) to predict the badness of Baltin’s there patttern. For in the environments where (48) forces raising of there to leave a lower resumptive pronoun, (62) in effect requires that pronoun to be controlled, while (68a) in effect forbids that, since the required controller cannot then serve as such. While this account is entirely informal, it seems basically coherent. Turn next to idiom chunks and proverbial expressions. A variety of these seem to provide a great deal of support for a principle like (48), for a very large number of them involve DPs that occur in antipronominal contexts. Some of these take inherent subjects and some have nonsubject antipronominal contexts that are passivizable and then form subject antipronominal contexts And remarkably, such expressions reveal all the types of restrictions on raising to object and raising to subject I have documented for nonidiomatic antipronominal contexts. A typical example is seen in (70):

128 (70)


a. b. c. d.

(She said) [birds of a feather]1 flock together (*and they1 do flock together). She believes birds of a feather to flock together. *She believes t1 to flock together [birds of a feather]1. *Sally may believe t1 to flock together and Sonia certainly believes t1 to flock together [birds of a feather]1. e. Birds of a feather are likely to flock together. f. *How likely t1 to flock together are [birds of a feather]1? g. *Likely t1 to flock together though [birds of a feather]1 are, . . .

There is a problem relevant to testing, for example, principle (35) against such idiomatic data, which is common though not severe in (70). This involves the fact that most often the relevant idiomatic nominal that appears in an antipronominal context is not particularly ‘heavy’ in the sense in which this might be a requirement for Complex DP Shift and certainly a preference for Right Node Raising. But in all cases, one can see that the effect in analogs of (70c and d) cannot be reduced to violation of any ‘heaviness’ requirement. Consider for instance (71): (71)

a. The cat is out of the bag. b. Sonia believes the cat to be out of the bag. c. They said the cat was out of the bag and it is out of the bag. (ok literal/*idiomatic)

Both (71a and b) are ambiguous, having either a literal reading about a feline or one that involves a secret. But not so in (72): (72)

a. *Sonia believes t1 to be out of the bag [the cat]1. b. ?Sonia may believe t1 to be out of the bag and Harriet certainly does believe t1 to be out of the bag [the cat]1.

Here there may be violations of heaviness constraints on the pivot nominals for Complex DP Shift and Right Node Raising, although I believe that heavy stress on the pivot renders it acceptable. But independently of that, it is palpable that only the feline reading is possible, which cannot have anything to do with heaviness, since the pivot is equally heavy under either reading. Consequently, some principle must block the idiomatic reading but not the literal one, and a principle that forces a resumptive pronoun combines with the antipronominal character of the subject position of the idiom to play this role. One can cite a great deal more evidence that supports (48) based on subject antipronominal contexts linked to idiomatic or proverbial nominals. I have given a range of such data in the most restricted form possible. In all cases, strictures about the irrelevance of heaviness factors already touched on should be borne in mind. Moreover, all stars represent only idiomatic or proverbial readings. The data is ordered essentially randomly in (73)–(78): (73)

a. [A great deal of attention]1 was claimed to have been paid to that (*but it1 wasn’t paid to that).


129

b. *Ernest believes t1 to have been paid to that [a great deal of attention]1. c. *How likely t1 to have been paid to that was [a great deal of attention]1? (74)

a. Sabine said [the chickens]1 had come home to roost (*but they1 hadn’t come home to roost). b. *Sabine may have found t1 to have come home to roost and Emily certainly found t1 to have come home to roost [the chickens]1. c. *Likely t1 to have come home to roost though [the chickens]1 were, . . .

(75)

a. Gwen claims [Tony’s heart]1 is in the right place (*and it1 is in the right place). b. *Gwen may have believed t1 be in the right place and Nora certainly did believe t1 to be in the right place [Tony’s heart]1. c. *How likely t1 to be in the right place was [Tony’s heart]1?

(76)

a. [Cold water]1 was thrown on your idea (*but it1 wasn’t thrown on my idea). b. *Dana may believe t1 to have been thrown on your idea and she certainly believes t1 to have been thrown on my idea [cold water]1. c. *Likely t1 to have been thrown on his idea though [cold water]1 was, . . .

(77)

a. Myra believes [the early bird]1 gets the worm (*but it1 doesn’t get the worm). b. *Myra believes t1 to get the worm [the early bird]1. c. *How likely t1 to get the worm is [the early bird]1?

(78)

a. [The rug]1 was pulled out from under him (*but it1 wasn’t pulled out from under me). b. *Andrea believes t1 to have been pulled out from under him [the rug]1. c. *Likely t1 to have been pulled out from under him though [the rug]1 is, . . .

I must leave it to the reader to determine that entirely parallel data can be found for such further expressions as those in (79): (79)

a. The ball is in your court. b. A stitch in time saves nine. c. All hell broke loose. d. A good time was had by all. e. The fat is in the fire. f. The cat has his tongue. g. The lovebug bit Tony. h. The shoe is on the other foot. i. Fortune smiled on Gwendolyn. j. The bottom fell out of the cocoa market. k. The buck stops here. l. The jury is still out on that proposal. m. The shit hit the fan. n. The jig is up. o. The party is over. p. The good times are about to roll.

130


q. The worm is going to turn. r. The tide is turning. s. All the ducks are in line. t. The truth will out. Murder will out. u. The wind has gone out of his sails. v. The fox is in the chicken coop. w. The wolf is at the door. x. The ice was broken at the party. y. The devil alone knows who did this. z. His number is up. z1. Miriam’s ship came in. z2. A pall fell over the gathering. z3. The baby was thrown out with the bathwater. z4. Close tabs were kept on her movements. z5. Unfair advantage was taken of his good nature. z6. Stan’s goose is cooked. z7. Churchill’s shoes will be difficult to fill. z8. Strings were pulled. z9. That filthy habit was finally kicked. z10. Simone’s hair stood on end. z11. The shoe is on the other foot. z12. The ceiling caved in on Mike. z13. A damper was put on the evening by Tony’s announcement.

So far then, the idiomatic and proverbial expression data just seem to instantiate multiple further instances of antipronominal contexts, all of which bar raising to object and raising to subject under the same conditions already uncovered for other types of antipronominal context. All these data thus initially appear only to strengthen the support for a principle like (48) forcing resumptive pronouns in certain, but only certain, instances of raising. However, a range of partially different data that involve the same idiomatic and proverbial expressions may call this conclusion into question. It can be illustrated by the facts in (80): (80)

a. He kept close tabs on her movements. b. *He believes t1 to have been kept on her movements [very close tabs]1. c. *Ted may have believed t1 to have been kept on her movements and Fred does believe t1 to have been kept on her movements [very close tabs]1. d. *He kept t1 on her movements [very close tabs]1.

Hitherto, the data given relevant to the interaction of idiomatic DPs and Complex DP Shift or Right Node Raising was only of the types in (80b and c), where the phrase extracted to the right is a raisee. But (80d) illustrates something that turns out to be equally systematically true, namely (81): (81)

No idiomatic/proverbial DPx can be the extractee for Complex DP Shift or Right Node Raising even when DPx is not a raisee.


131

Further supporting evidence for (81) is given in (82): (82)

a. Helen let the cat out of the bag. b. Helen let t1 out of the bag [the cat]1. (ok literal/*idiomatic) c. Helen threw cold water on Ted’s face/idea. d. Helen threw t1 on Ted’s face/*idea [cold water]1. e. Helen pulled the rug out from under Ted. f. Helen pulled t1 out from under Ted [the rug]1. (ok literal/*idiomatic) g. No one can fill Ted’s shoes in this organization. h. *No one can fill t1 in this organization [Ted’s shoes]1. i. Helen threw the baby out with the bathwater. j. Helen threw out the baby with the bathwater. k. *Helen threw t1 out with the bathwater [the baby]1.

The potential problem that generalization (81) raises is this: If there is no way to reduce it to principle (48), then some other principle must block the bad cases of (80) and (82). And this principle could then arguably be invoked also for the idiomatic data that seem to support (48), like (80b and c), undermining some of the support for (48). And so far I see no way to reduce (81) to (48) or any way to guarantee that the bad cases of (82) would contain resumptive pronouns in the slots that are arguably antipronominal. For the key to guaranteeing their presence in earlier cases was to invoke conditions under which raisings are required to link to resumptives, and in (82) there are, arguably, no raisings. Moreover, there is a second class of data that involves idiomatic expressions illustrated by (83): (83)

a. b. c. d. e. f. g. h.

They said she threw cold water on his head and she did throw it on his head. *They said she threw cold water on his idea and she did throw it on his idea. ?They said cold water was thrown on his head and cold water was. *They said cold water was thrown on his idea and cold water was. They said cold water was thrown on his head it was thrown on his head. *They said cold water was thrown on his idea and it was thrown on his idea. They said cold water was thrown on his head and it was. They said cold water was thrown on his idea and it was.

It seems that most idiomatic DPs that can appear as subjects reveal a patttern like that seen in (83f and h). While the subject is antipronominal when another piece of the idiom is present in the surface form, the position allows a weak pronoun when no piece of the idiom appears there. Now one might generalize these observations by suggesting that there is a principle for idioms composed of multiple lexical pieces something along the lines of (84): (84)

The Multiple Lexical Item Idiom Condition If a multiple-membered set of lexical items K forms an idiom and one member of K has a visible surface realization, then every other member of K must have a surface realization and all these surface realizations form a c-command chain (most accurately, one that involves a version of c-command that ignores prepositions, as in Pesetsky [1995: 172–3]; see for example, [82i, j]).

132


Such a principle cannot only block data like (83f and h); it also blocks the bad instances of (82) under the well-established assumption that the DP victim of Complex DP Shift ends up in a structure something like (85): (85)

[ [Verb, X, t1,Y] + DP1]

The problem then is that (84) not only covers the cases of idiomatic blockages that (48) might, but it also covers those where (48) is irrelevant. Principle (84) thus seems to undermine any support for the earlier analysis drawn from the idiomatic domain. This does not, of course, yield grounds for rejecting the overall analysis but does weaken its factual support somewhat. Notice that (84) properly is consistent with pieces of idioms being passivized DPs or DPs that raise under the construction in (55)/(58) or under object raising, for in these cases a c-command chain is maintained. Similarly, (84) is also consistent with pre- or postparticle word order for idiomatic DPs, as in (82i, j).

8. An earlier proposal One should compare this proposal with that of Lasnik and Saito (1992), with which it shares key properties but from which it differs in key ways. (86)

a. *How likely t1 to be a riot is there1? b. There is likely to be a riot.

(87)

a. *How likely t1 to be taken of John is [advantage]1? b. Advantage is likely to be taken of John.

(88)

a. How likely t1 to win is [John]1. b. John is likely to win.

Recall that these authors proposed to account for (86a), (87a), and related data via principle (67), which requires that a trace be bound (hence c-commanded) by its antecedent, “throughout a derivation.” As these authors note, (67) raises a problem for standard GB analyses of a case like (66a), repeated as (89): (89)

How likely t to win is John?

For (67) determines that (66a) cannot actually have structure (89), in which the trace would not be c-commanded by John at S-structure. Lasnik and Saito then propose that traditional raising to subject structures like (66b) have dual analyses along the lines of (90a, b): (90)

a. John is likely t to win. b. John is likely PRO to win.

Example (66a) can then be taken to instantiate (91b), which satisfies principle (67), and not the blocked (91a):


(91)

133

a. How likely t to win is John? b. How likely PRO to win is John?

In these terms, Lasnik and Saito (1992) claim that (86a) and (87a) are ill formed because of the known general failure (but see [68b]) of, for example, expletives like there and idiomatic NPs like advantage to function as controllers, that is, immedate antecedents of PRO structures, which give data like (92): (92)

a. *There tried PRO to be a riot. b. *Advantage wants PRO to be taken of John.

Similarities with the view I have advocated here are thus clear, in that Lasnik and Saito also invoked a type of control. But unlike my proposal, which involves controlled resumptive pronouns, theirs involves the nonresumptive pronoun PRO, in effect first introduced in Postal (1970) as a pronoun with a Doom feature. However, explication of the deviance of (86a) and (87a) via principle (67) and the proposal that traditional raising structures actually have dual movement/control analyses were certainly not tenable claims in Lasnik and Saito’s terms. Unless buttressed by unknown and unstated additions, the proposals block many perfectly grammatical structures. Most strikingly, Lasnik and Saito’s analysis undermines the adequacy of GB and, more generally, transformational movement/trace analyses of passives, which they nonetheless endorsed; see their page 127. For under the standard analysis (93a and b) violate (67) as much as, for example, (89) does: (93)

a. Shocked t1 by the revelations though (they claim that) [Arthur]1 was, . . . b. They said [he]1 would be eaten by the shark and eaten t1 by the shark (I suspect that) [he]1 was.

To maintain (67), Lasnik and Saito would thus have to provide dual analyses for passives, with the novel analysis not involving an object position trace. In this case, PRO cannot provide the second analysis since, as the authors stress (1992: 130), they maintain the earlier GB idea that PRO cannot be governed. So (94b) is not a possible analysis for (94a): (94)

a. He was eaten by the shark. b. He was eaten PRO by the shark.

It remains entirely obscure what, if any, analysis distinct from the standard trace one they actually maintained Lasnik and Saito could give for passives so as to allow cases like (93). The problem just raised for principle (67) as the basis of the subject raising data I have gone over is, evidently, not inherently linked to passives. Clearly it arises in any case where there is the possibility of the left-extraction of a constituent that contains what in GB terms would have to be a subject trace. If, as has often been proposed, middles, unaccusatives, and such, are of this type, then (95a and b), for example, pose the same difficulty that (94a and b) do:

134 (95)


a. They said [the convertible]1 handled well and handle t1 well (I am sure that) [it]1 did. b. They said that [that]1 would happen and happen t1 (they claim that) [it]1 did.

A distinct problem for Lasnik and Saito’s proposal is, of course, that it has no way to account for the antipronominal context data. Recall that the crucial point there was that raisings can either involve resumptive pronouns or not, and violations ensue in antipronominal contexts only under those circumstances where such pronouns are forced. Consider (96): (96)

a. Helen believes something exactly like that/*it to have been the matter with the transmission. b. Helen believes something exactly like that/it to have been harmful to the transmission. c. *Helen believes t1 to have been the matter with the transmission [something exactly like that]1. d. Helen believes t1 to have been harmful to the transmission [something exactly like that]1.

There is nothing in Lasnik and Saito’s proposal that accounts for contrasts like (96c and d), nothing to link this contrast to the pronoun contrast between (96a and b). Even if in some unknown way they could force PRO in (96c), that would not suffice to block the example, for unlike the expletive and idiomatic data they dealt with, a DP like the pivot in (96a) is a perfectly happy controller of what they would have to take as PRO, as in (97): (97)

[Something very grave]1 can go wrong on one day without PRO1 going wrong on another day.

Moreover, they would have no way to account for (98): (98)

*[Something very grave]1 can be wrong with one car without PRO being wrong with another.

In our terms, (98) is of course bad because control involves pronouns, and the controlled subject of be wrong with (but not that of go wrong) is an antipronominal context. Return to Lasnik and Saito’s claim that raising to subject structures have dual analyses and that, for example, (91a) can have a structure like (91b). This is dubious on the basis of the two novel tests for the control/raising distinction introduced in section 5.3 of chapter 2. Recall that it was argued that control complements cannot be middles and cannot be the sort of stock (price) designating metonymy structures represented by company names. These constraints yield control/raising contrasts like (99): (99)

a. Control: *Uruguayan police want to bribe cheaply. b. Raising: Uruguayan police seem to bribe cheaply.


135

c. Control: *Microsoft wanted to go up. d. Raising: Microsoft seemed to go up.

These factors independently distinguish what are differentiated as raising (that is, trace in Lasnik and Saito’s terms) versus control (PRO) structures in GB terms. If these usages are good tests for the raising/control distinction, Lasnik and Saito’s principle (67) account of (91) would then predict that there can be no left extracted cases parallel to (91) with middles or the ‘stock’ usage in question, since that for them must instantiate PRO and not trace. But the facts are otherwise, as (100) and (101) show: (100)

a. How likely t1 to go up is [Microsoft]1? b. They said Microsoft was likely to go up and likely t1 to go up [Microsoft]1 still is. c. Likely t1 to go up though [Microsoft]1 is, . . .

(101)

a. How likely t1 to bribe cheaply are [Uruguayan police]1? b. They said Uruguayan police were likely to bribe cheaply and likely t1 to bribe cheaply [Uruguayan police]1 still are. c. Likely t1 to bribe cheaply though [Uruguayan police]1 are, . . .

9. A speculation Before concluding this chapter, so far restricted to facts from English, I would like to make one rather speculative suggestion of crosslinguistic relevance. A number of NLs, which include Swedish and the West African languages Vata and Yoruba, manifest an interesting property linked to resumptive pronouns. For the facts in the latter, see Carstens (1985a, 1985b), Koopman (1982, 1984), Koopman and Sportiche (1982/1983, 1986), and Stahlke (1974). I will limit my remarks to Swedish. Engdahl (1985, 1986), Engdahl and Ejerhed (1982), Zaenen and Maling (1982), and others observe that Swedish, in contrast to English and even to other Scandanavian languages, does not permit gap-yielding extraction of the subject of an embedded clause with lexical material preceding the subject. Rather, a visible resumptive pronoun must appear, as in the minimal pair in (102): (102)

Zaenen and Maling’s (1982) (5a, b) Vem1 undrade alla om *t1/han1 skulle komma i tid? Who1 wondered all if null1/he1 would come on time “Who did everyone wonder if he would come on time”

Let us refer to resumptive pronouns of this sort as b-resumptive pronouns. The relevant works reveal that b-resumptive pronouns like that in the good version of (102) have a number of characteristic properties; specifically, they cannot alternate with epithets, can participate in across-the-board extractions as if they were gaps, can serve as the licensing gaps for parasitic gaps, and do not suppress weak

136


crossover effects, a property that has been noted to be associated with a range of resumptive pronouns by Cinque (1990), Lasnik and Stowell (1991), May (1985), and Postal (1993b). Moreover, b-resumptive pronouns only appear as subjects. In all these respects, b-resumptive pronouns seem to contrast with other Swedish resumptive pronouns and to behave essentially like ordinary extraction gaps; that is, they behave as if they were not there. The analysis of the puzzling restrictions on English Complex DP Shift and Right Node Raising in connection with raising of subjects out of antipronominal contexts, namely, one that involves conditional determination of resumptive pronouns by the raising, not the extractions, suggests the possibility of a parallel account of Swedish b-resumptive pronouns. Namely, suppose structures like (102) involve ordinary subject extractions that yield ordinary gaps, while the resumptive pronouns are determined conditionally by some kind of raising associated only with subjects. The rule would then be that the raising in question yields a b-resumptive pronoun if and only if the raisee is extracted. To make this work, it would be necessary to regard the preverbal subject position as one that involves raising, which is not today all that controversial. However, clauses that contain Swedish b-resumptive pronouns would have to involve two raisings, first of the nonresumptive subject, subsequently extracted, and second raising of the b-resumptive. This issue does not arise in the English cases I have discussed because in these the resumptives are invisible. Working out the needed analysis of Swedish is then not without problems but does not seem beyond feasability.

10. Conclusion I conclude by briefly making four points. First, I believe I have given a new and solid argument for raising to object analyses based on shared properties of raising to object and uncontroversial raising to subject cases. The argument depends on principle (48), which conditionally determines the presence of resumptive pronouns in raising cases, accounting for gaps in both types of raising paradigms where subject antipronominal contexts are found. Second, the discussion has strengthened the evidence for the recognition of invisible pronouns in topicalization, cleft, object raising, object deletion, and P-gap structures by showing that this assumption plays a key role in accounting for the special type of constraints that is the topic of this essay. Third, while there are similarities between what is suggested here and the proposal of Lasnik and Saito (1992), the proposal here has been argued to have several clear advantages. Fourth, of considerable theoretical significance is the nature of the constraint found to link the requirement of a resumptive pronoun for raisings to phenomena that bring it about that the raisee follows the raising constituent remnant. While such a constraint is easily stated in a framework of nongenerative, statement-based description that appeals to a view of sentences as built of arcs and their relations, as in (i) of note 2, the condition would have what I earlier called a ‘nonderivational’, ‘noncyclic’ character if one attempted to express it internal to a proof-theoretic/deri-


137

vational view of grammars. Put simply, in the latter terms one would have to say that the choice of whether a resumptive pronoun is required in the complement is determined nondetermininistically by whether those grammatical phenomena that determine the form of the main clause yield a word order in which the remnant precedes the raisee or not. This yields a sort of necessary ‘look-ahead’ feature. If the claim that prooftheoretic operations must apply in a cyclic fashion, often taken to be an important principle, has content, ‘look-ahead’ requiring derivations should be impossible. Cases like those central to this chapter therefore seem to be clear counterexamples and represent by themselves a sharp challenge to such ideas.6

138


4

Chromaticity An Overlooked English Grammatical Category Distinction

1. Basics This essay argues for the existence of an English grammatical category distinction that, as far as I know, has not been previously recognized. This distinction bifurcates the class of nominals into subtypes I will refer to as chromatic and nonchromatic. Instances of these are seen in (1) and (2): (1)

Chromatic DPs = {some fox, any fox, no fox, what fox, whatever fox, many computers, that hat, much soup, some place, some week, what way, what reason, . . .}

(2)

Nonchromatic DPs = {something, anything, nothing, what, whatever, squat, stuff, someone, everyone, who, somewhere, where, sometime, when, how, why . . .}

Most DPs belong to the chromatic class, which is open. The nonchromatic class consists, though, of a restricted, if not tiny, group of forms.1 The choice of the terminology adopted is based on the fact that systematically the nonchromatic forms have extremely general and nonspecific meanings, ones, as it were, that lack ‘color’. Kishimoto (2000: 563) approaches the same meaning property by speaking of the nouns that underlie nonchromatic DPs as “semantically light” and claiming they are “devoid of lexical meaning.” The latter claim seems too strong. So, while both of the objects in (3) are nonchromatic, certainly they differ in meaning:2 138

CHROMATICITY

(3)

139

a. Rhonda criticized something unusually evil. b. Rhonda criticized someone unusually evil.

That is, the DP in (3b) restricts the object referent to humans, which that in (3a) does not. If neither had any lexical meaning, how could they differ? A proper subset of nonchromatic DPs is represented by the elements that Kishimoto (2000) argues involve nominal-internal N-raising into the associated D. This N-raising can be taken to underlie at least two properties of the relevant subset, the characterization of which is briefly touched on in section 5. (4)

a. Unlike chromatic DPs, unrestricted in this regard, elements of the subset are systematically single-word forms. b. Those nonchromatic DPs that have property (4a) never permit prenominal simplex adjectives.

Following Kishimoto’s logic, which I believe is essentially correct, this would follow from a structure for the forms in question along the lines of (5):3 (5)

[DP [D Dx + N] [NP (Adj) [Noun Ø ] ] ]

That is, since the normal position of simplex adjective phrases is prenominal within NP, if the N ends up outside the NP but inside the D it will precede any simplex adjectives. As Kishimoto (2000) observes then, this yields rightly contrasts like (6): (6)

a. b. c. d.

some vicious fox *some fox vicious something vicious *some vicious thing

Actually, (6d) is not, of course, in fact ungrammatical. But I would argue that on its good analysis, it represents a chromatic DP; thus a nominal stem thing is found in both the chromatic and nonchromatic categories (whereas, e.g., the stem fox is found only in the former). This dual view of certain N-stems like thing is also advocated by Kishimoto (2000) and is supported further later. While it is argued here that the chromatic/nonchromatic distinction is a syntactic one, it seems to correlate with some kind of semantic contrast that underlies the choice of terminology. This involves a lack of whatever semantic substance it is that a typical lexical noun contributes. The property shows up clearly in the case of pseudoclefts, as in (7): (7)

a. b. c. d. e. f.

What Edgar purchased was some house/boat/manuscript. *What Edgar purchased was something. What Edgar purchased was not any house/boat/manuscript. *What Edgar purchased was not anything. What Edgar purchased was something interesting. What Edgar purchased was not anything that I would have wanted.

140


That is, the predicate in a pseudocleft cannot consist of a simple existential nonchromatic DP. It can, however, consist of that plus a modifier, as in (7e, f). The restriction might seem to be that the predicate DP add some semantic substance. If so, the contrast in (8) reveals something important about the ‘semantically light’ or ‘nonchromatic’ aspect of the semantics of nonchromatic nouns. That is, while the clearly syntactically nonchromatic form someone is animate and human like the chromatic a person, the former seemingly lacks a semantic property that the chromatic a person has: (8)

What Edgar saw was *someone/a person.

Perhaps one can appeal to a difference between presupposed or backgrounded meaning and another kind, the former being all that can be associated with nonchromatic nouns. But this topic is beyond the scope of these remarks. So far three properties that systematically distinguish (a subset of) nonchromatic DPs from more common chromatic DPs have been noted: (i) single-word status, (ii) nominal-internal position of simplex adjective phrases, and (iii) possibility of occurrence as focus of pseudoclefts. A fourth property involves the possibility of combination with pre-D, prenominal modifiers, in particular, even and only. These are only possible with chromatic DPs (here capitalization indicates strong stress): (9)

a. b. c. d.

Even every MAN agreed with that. *Even everyone agreed with that. Even some GORILLA spoke Spanish. *Even someone spoke Spanish.

(10)

a. b. c. d.

Only every NURSE advocated such a policy. *Only everyone advocated such a policy. Only some CHIMP spoke Spanish. *Only someone spoke Spanish.4

There is, arguably, also a fifth property sensitive to the chromatic nonchromatic DP distinction. This involves the distribution of the form else: (11)

a. b. c. d. e. f. g. h. i. j. k.

Herb discussed every/some/no problem/issue (*else). Herb discussed everything/something/nothing (else).5 Herb looked in every/some/no place/bar (*else). Herb looked everywhere/somewhere/nowhere (else). Did Ethel check any room (*else)? Did Ethel check anything (else)? What (*else) bar (*else) did they burn down? What (else) did they burn down? Whatever (*else) principle (*else) we adopt, they will reject it. Whatever (else) we adopt, they will reject it. Some (*else) linguist (*else) was rejected.

CHROMATICITY

141

l. Someone (else) was rejected. m. What (*else) linguist (*else) was rejected? n. Who (else) was rejected?

It appears, that is, that else is possible (only) with members of the same subset of nonchromatic DPs that is characterized in (4).6 Further, consider the issue of which DPs can occur as preposed genitives: (12)

a. b. c. d.

some/every/no doctor’s wife someone’s/everyone’s/no one’s wife what/which man’s wife whose/who else’s wife

With human DPs, preposed genitives are clearly not restricted as to chromaticity. But with nonhuman DPs, contrasts appear: (13)

a. b. c. d.

some/every/no car’s carburetor *something’s/everything’s/nothing’s carburetor what/which car’s carburetor *what’s/*what else’s carburetor

(14)

a. b. c. d.

some/every/not a single/any person’s blood someone’s/everyone’s/not a single one’s/anyone’s blood what/which person’s blood whose/who else’s blood

(15)

a. b. c. d.

some/every/not a single/any car’s oil *something’s/*everything’s/*not a single thing’s/*anything’s oil what/which car’s oil *what’s/*what else’s/*whose (inanimate) oil

(16)

a. that car’s motor b. *that’s motor

The initial generalization seems to be that for inanimates, genitives are systematically possible on chromatics but blocked on nonchromatics. There is, though, one problem. While whose cannot of course be used as an inanimate interrogative form, it is a fine inanimate relative pronoun: (17)

a. an integer whose successor is even b. an/some explosion whose cause was unknown c. no discovery whose origin is at issue

The generalization proposed would then lead one to expect that nonchromatic heads with such relative pronouns would be impossible. But this does not seem to be in accord with the facts:

142 (18)


a. something whose analysis is incomplete b. everything whose cause is unknown c. nothing whose origin is at issue

It is argued in section 3.5 that restrictive relative pronouns agree with their heads in chromaticity. If so, then the instances of whose in (18) would all be nonchromatic and the generalization that inanimate genitive nonchromatics are blocked would be false. I have no better solution at this point than to say that the sixth generalization, that which links nonchromatics and inanimate genitives, simply does not hold for relative pronouns. Since it does hold elsewhere, though, an argument for the relevance of the chromaticity dimension in English syntax still emerges. A seventh grammatical generalization also supports recognition of the chromaticity dimension. This involves a restriction on topicalization illustrated in (19): (19)

a. Jerome understands something/everything/someone/everyone/stuff. b. Jerome understands some movies/every movie/some singers/every singer/cheap stuff. c. *Something/*Everything/*Someone/*Everyone/*Stuff, Jerome understands. d. Some movies/Every movie/Some singers/Every singer/Cheap stuff, Jerome understands.

The generalization, in effect partly noted in Postal (1993c), appears to be that ‘unmodified’ nonchromatic DPs, those that consist at most of a combined D + N, cannot be topics.7

2. Selections So far, seven properties have been cited that support the existence of a chromatic DP/nonchromatic DP distinction. However, the most important features that argue for the existence of the grammatical categories of chromatic and nonchromatic DPs involve selections. While standard contexts that permit DPs make no selectional distinction between chromatic and nonchromatic DPs, as in (20), there is, nonetheless, a range of environments that distinguish chromatic and nonchromatic DPs: (20)

a. Henrietta cooked some fish/something. b. The priests offered nothing/no gifts to the demons.

First, various environments permit DPs of the form some/any/no/the, and so on + lexical noun but bar all of the inanimate indefinite pronouns something, anything, and nothing, their wh variants what, whatever, and so on. Examples are given in (21)–(28): (21)

a. Joe attended some/no/the Catholic school/*something/*nothing/*stuff/*that thing. b. Joe didn’t attend any school/*anything/*squat. c. *Whatever he attended was in Boston.

CHROMATICITY

143

(22)

a. Sue spoke some/every Slavic language/the language/*the thing/*something/*everything/*stuff. b. Sue didn’t speak any Slavic language/*anything/*squat. c. *What (else) do they speak in Bangladesh?

(23)

a. Mike scored some baskets/the basket/*the thing/*something/*stuff. b. Mike didn’t score any baskets/*anything/*squat. c. *Whatever Mike scored was not a three pointer.

(24)

a. Mike took some time/the time/*something/*everything/*stuff/*the thing to call his mother. b. Mike didn’t take any time/*anything/*squat to call his mother. c. What *(amount of time) did Mike take to mark the exams?

(25)

a. b. c. d.

Vanessa committed perjury/*something/*everything/*nothing. What crime/*What did Vanessa commit? No matter what crime/*what Vanessa committed, I still respect her. Vanessa didn’t commit any crime/*anything.

(26)

a. b. c. d.

Glen swore an oath/*something/*everything/*nothing. What kind of oath/*What did Glen swear? Whatever oath/*Whatever Glen swore, he was just kidding. Glen didn’t swear any oath/*anything/*a damn thing.

(27)

a. b. c. d.

Vanessa carried out a threat/some threat/*something. What threat/*What (else) did Vanessa carry out? No matter what threat/*what Vanessa carried out, . . . Vanessa didn’t carry out a single threat/any threat/*anything/*a damn thing.

(28)

a. b. c. d.

Kim spent some part/*something of her life in Topeka. Kim didn’t spend any part/*anything of her life in Topeka. What part/*What of her life did Kim spend in Topeka? Whatever part/*Whatever (else) of her life Kim spent in Topeka, . . .

A further case is pointed out by my colleague Mark Baltin: (29)

a. Arnie dissuaded/deterred Rosalie from some course of action/*something/*everything/*stuff. b. Arnie did not dissuade/deter Rosalie from any course of action/*anything/*a damn thing/*squat. c. What course of action/*What did Arnie dissuade/deter Rosalie from? d. No matter what course of action/*what Arnie dissuaded/deterred Rosalie from, . . .

One can, in the light of earlier remarks, characterize the contexts in (21)–(29) as ones that obligatorily require chromatic DPs or, equivalently, ones that exclude nonchromatic DPs.

144


Note that the chromatic/nonchromatic distinction may cleave differently for a morphological verb depending on its meanings. That is, one meaning may require a chromatic type of DP in a particular context, another meaning not, as illustrated in (30): (30)

a. b. c. d. e. f.

He didn’t spend <monetary> any money/anything/squat in Hawaii. He didn’t spend any time/*anything/*squat in Hawaii. He didn’t assume any principle/anything/squat. He didn’t assume any position/*anything/*squat. He didn’t attend <matriculation> any school/*anything/*squat. He didn’t attend any lecture/anything/squat at that conference.

So spend in the monetary sense is indifferent to the chromaticity of its object, but spend in the temporal sense requires a chromatic object. Similarly, assume in the logical sense takes either chromatic or nonchromatic objects but in the physical sense requires a chromatic one. The same pattern is seen with attend, whose physical sense is equally compatible with object chromaticity or its absence but whose matriculation sense requires a chromatic object. Such facts seem quite anomalous at first glance but have at least a simple description if one can appeal to the grammatical category distinction argued for here. A second category of selectional fact that supports chromaticity as a grammatical category dimension in English is that some DP environments require nonchromatic DPs. Two such environments are formed by the subjects of the expressions be the matter with + DP and be wrong with + DP, noted to require antipronominal subjects in chapter 3: (31)

a. Nothing/Something/Everything/(Terrible) Stuff is the matter/wrong with that car. b. *Some problem/*Power loss/*Rust/*Old age/*Overuse is the matter/wrong with that car. c. I don’t believe anything/squat/a fucking thing to be the matter/wrong with that car. d. *I don’t believe a flat/rust/an oil leak/electrical problems to be the matter/wrong with that car.

Another nonchromatic environment, whose existence I am indebted to Christopher Potts for pointing out to me, is the subject of the use of eat on which it means something like “bother”: (32)

a. b. c. d.

Something/(Bad) Stuff/Nothing/Not a damn thing/That (thing) is eating Gilbert. What (else) is eating Gilbert? *Some problem/Hair loss/Boredom/Lack of charisma/Overweight is eating Gilbert. I don’t believe anything/squat/jack-shit to be eating Gilbert.

A further environment that selects nonchromatic DPs is seen in (33): (33)

a. They named their daughter something/that/the same thing I named mine/*some (stupid) name.

CHROMATICITY

145

b. They did not name their daughter anything/a damn thing. c. I need to know who named their daughter what (*name).

And further adverbial instances are found in (34) and (35): (34)

a. b. c. d. e.

Trolls resided *some bridge/*every bridge/somewhere/everywhere. Trolls resided *no bridge/nowhere. Trolls didn’t reside *any bridge/anywhere. *What bridge/Where (else) did trolls reside? No matter *what bridge/where trolls resided, . . .

(35)

a. Juanita played golf *some way/*some manner during her vacation. b. *What way/*What manner/How (else) did Juanita play golf during her vacation? c. No matter *what way/*what manner/how (else) Juanita played golf, . . .

Taking it for granted now that there are selections defined on the chromatic/ nonchromatic distinction, the question arises of whether all such facts can be taken to involve purely semantic requirements of the selectors in question. This question is relevant, evidently, to the issue of whether all selection is semantic; see, for example, Grimshaw (1979, 1981) and Pesetsky (1995). Consider first a verb like attend on its matriculation sense, which requires a chromatic object. Could this follow from some purely semantic requirement? It is not easy to see how. The relevant requirement would seem to be (36): (36) The object of matriculation attend denotes an institution of learning.

The problem then is that there is no reason that such a condition would block, for example, (37a and b), given that, as (37c and d) illustrate, there is no general barrier to nonchromatic forms denoting institutions of learning: (37)

a. b. c. d.

*Celia attended <matriculation> something/everything/stuff. *What Celia attended <matriculation> was the University of Vermont. Celia was discussing something (namely, her high school) with Billy. What Celia was referring to was the University of Vermont.

Given data like (37c and d), there is no visible reason that the nonetheless clearly deviant (37a and b) would violate what appears to be the semantic condition on the relevant verb, matriculation attend, that represented as (36). Turning to the opposite sort of selectors, those like matter with, which require nonchromatic subjects, it is also difficult to envisage a motivated purely semantic condition that would block only the bad cases of (38a), especially given the grammaticality of those like (38b): (38)

a. They didn’t prove anything/a damn thing/squat/*leukemia/*bulimia/*whooping cough to be wrong with her. b. What they proved to be wrong with her was leukemia/bulimia/whooping cough.

146


My tentative conclusion then is that the selectional constraints that involve the chromaticity dimension are syntactic, entailing more generally that not all selection is semantic. To conclude this section, the two types of selectional evidence contribute, strongly, I think, to supporting the syntactic reality of the chromatic/nonchromatic distinction. But the sort of evidence that shows this perhaps most strongly of all involves what will be argued to be a range of agreement phenomena.

3. Chromatic agreement 3.1. Remark I believe that several types of constructions in which constituents of type A are known to manifest agreement with others of type B in properties such as person, gender, and number are such that constituents of type A must agree in chromaticity value with those of type B. This state of affairs is harder to show than for agreement that involves person, gender, or number, because what I suggest is chromaticity value agreement has no direct morphological realization. That is, no expression B that can be argued to manifest agreement in, say, positive chromaticity value differs in morphological form from the corresponding element that manifests negative chromaticity value. Nonetheless, there are restrictions that seem best characterized as agreement of chromaticity value. 3.2. Chromatic agreement of ordinary visible definite pronouns First, I suggest that ordinary definite pronouns like it must agree with their antecedents in chromaticity value. So, consider (39): (39)

a. b. c. d.

*Jerome spent something (namely, Easter) in Bermuda. Jerome spent some holiday (namely, Easter) in Bermuda. Jerome discussed something (namely, football/bunions/ice cream). Marsha discussed *something1/some holiday1 because Jerome spent Bermuda.

it1 in

The already mentioned fact that the temporal spend excludes nonchromatic objects accounts for the badness of (39a). But (39c) shows that discuss, like most verbs, is not subject to such a constraint. Nonetheless, the nonchromatic version of (39d) is ill formed. Given that the chromatic version is good, this cannot be due to the mere presence of a pronoun in the bad version. The facts would, though, follow if something like (40a and b) were both true: (40)

a. An English definite pronominal DP must agree in chromaticity value with its antecedent. b. The temporal verb spend requires any object DP it occurs with to be chromatic.

To make sense of (40a), one must assume minimally that the English definite pronominal forms it, its, they, them, their, and so on,8 are analyzed as both chromatic and nonchromatic. In this respect, they would be like the noun thing touched on ear-

CHROMATICITY

147

lier, which also needs to be taken to have both chromatic and nonchromatic forms, whereas most nouns are only chromatic. The point would then be that while, for example, it has syntactically distinct chromatic and nonchromatic variants, these do not differ morphologically in the way that, for example, the syntactically distinct singular masculine nominative and singular masculine and feminine forms he and she do. This makes chromaticity value agreement less visible than agreement in gender value but not necessarily thereby less real. Incidentally, the reason that all of the definite pronouns listed are inanimates has to do with the fact that while there are arguments (from else, etc.) that forms like someone are human nonchromatics, I do not know of any environment that selects for human chromatics or nonchromatics. Thus there is no environment currently available to show agreement in chromatic value for human DPs. Another case of the same sort as (39) is seen in (41): (41)

a. Mike spoke *something/some language. b. Mike studied *something1/some language1 despite not speaking it1. c. No matter *what1/what language1 Mike praised despite the fact that he did not speak it 1, . . .

Here the facts in (41b) would follow from (40a) together with assumption (42): (42)

The verb speak requires any object DP it occurs with to be chromatic.

A third case is found in (43): (43)

a. b. c. d.

Abdul composed his poems in some African language/*something/*everything. Abdul did not compose his poems in any African language/*anything/*a damn thing. What language/*What (the hell) did Abdul compose his poems in? Zeke was interested in some language1/*something1 because Abdul composed poems in it1. e. Whatever language1/*Whatever1 Zeke is interested in, Abdul composes a poem in it1.

A fourth case is (44): (44)

a. Sheila attended <matriculation> *something/some school. b. Sheila’s parents looked into *something1/some school1 because Sheila was thinking of attending <matriculation> it1 in the fall. c. No matter what 1/what school 1 Sheila’s parents praised, she refused to attend <matriculation> it 1.

And a fifth is: (45)

a. They diagrammed something (= physical position)/some physical position. b. They diagrammed *something1/some physical position1 right after Stan assumed it1. c. What1/*What position1 did they diagram right after Stan assumed it1?

148


A sixth case is: (46)

a. Their star scored *something/two goals. b. When that team needs *something1/a goal1, their star scores it1.

A seventh case is found in (47): (47)

a. b. c. d.

Barry drove in *something/some runs. *What did Barry drive in?9 Barry didn’t drive in *anything/*a damn thing/any run. The manager was hoping for *something1/a run1 just before Barry drove it1 in.

The strength of the latter case as support for principle (40a) is increased by the fact that the same verb also takes human objects with, strangely, pretty much the same meaning as the nonhuman objects of (47). But the human object position is not restricted with respect to chromaticity value. And, notably, analogs of (47d) are then fine: (48)

a. Barry drove in someone/a baserunner. b. She was pointing at someone1/a baserunner1 just before Barry drove him1 in.

I believe that the (arguably) agreement data just gone over support in a clear way the reality of chromaticity value agreement. And this seems true despite the lack of direct morphological manifestation of such agreement. Evidently, though, the existence of such a phenomenon is only possible on the basis of the underlying existence of the chromatic/nonchromatic category distinction itself. 3.3. Chromatic agreement and subject control The overall set of facts that relate to chromaticity agreement has nontrivial theoretical implications relevant to much-discussed issues. Consider first cases such as (49): (49)

a. Something1/A specific amount of money1 had to be discussed after Ø1 being spent <monetary> in Spain. b. *Something1/A specific week1 had to be discussed after Ø1 being spent in Spain.

The obvious factual point is that although in (49) there is no superficially visible pronoun to show agreement, the contrast in (49b) is nonetheless strikingly reminiscent of the chromaticity agreement contrast with temporal verb spend seen in (39), while the lack of contrast in (49a) parallels the lack of a chromaticity agreement contrast with the monetary verb spend seen in (50): (50)

Something1/A specific amount of money1 was set aside/saved so it1 could be spent <monetary> on vacations.

One finds in (49) of course an after adjunct with a controlled subject. Now, it has been an issue since at least the account in Postal (1970) whether cases of control

CHROMATICITY

149

involve invisible pronominal elements or not. The work just cited argued for the former view. But many have since denied it; see, for example, Chierchia (1984), Chierchia and Jacobson (1985), Jacobson (1992b), and Dowty (1985). What seems to be a chromaticity agreement fact in (49) then supports the invisible pronominal view. For such cases would force an alternative to the invisible pronominal view to somehow account for the contrast in (49) with some further mechanism M distinct from that in (40), where M would have nothing to do with pronominal agreement. To drive home the relevance of chromaticity constraints to control, it is important to see that (49) is not an isolated or anomalous instance of subject control. Relevantly then, control into subject complements also reveals parallels to clear pronominal chromaticity value agreement: (51)

a. *Something /A strange position was assumed by the gorilla. b. Its1 having been assumed by the notorious gorilla made *something1/a strange position1/famous. c. Ø1 Having been assumed by the notorious gorilla made *something1/a strange position1 famous.

(52)

a. *Something/Some language was spoken by the cloned scientist. b. Its1 having been spoken by the cloned scientist wouldn’t necessarily make *anything1/any language1 of interest to the CIA. c. Ø1 Having been spoken by the cloned scientist wouldn’t necessarily make *anything1/any language1 of interest to the CIA.

Chromatic relevance to subject control is also seen in structures with too/enough: (53)

a. Some school1/*Something1 had to be too expensive/cheap enough for it1 to be attended by Sheila. b. Some school1/*Something1 had to be too expensive/cheap enough Ø1 to be attended by Sheila.

The point is also made by the sort of subject control found in conjoined clausal structures with then: (54)

a. Some language1/*Something1 was chosen and then, Stella claims, it1 was spoken by all of the students. b. Some language1/*Something1 was chosen and then, Stella claims, Ø1 was spoken by all of the students.

Overall then, it seems clear that chromaticity facts argue for the role of invisible pronouns in a range of subject control structures. 3.4. Chromatic agreement and nonsubject control Chromaticity value agreement is arguably also involved in several distinct types of nonsubject control. First, there is the type associated with too/enough (see, e.g., Fiengo and Lasnik, 1974; and Postal, 1994b):

150 (55)


a. Some position1/*Something1 was too difficult for Jerome to assume b. Some position1/*Something1 was too difficult for Jerome to assume

it1. Ø 1.

Second, chromaticity is relevant to nonsubject control in those purposives of the sort that permit it (see, e.g., Bach, 1982; and Browning, 1987): (56)

a. b. c. d. e. f. g.

He composed poems in some Siberian language/*something. He chose *something1/some language1 for the purpose of composing poems in it1. He chose *something1/some language1 to compose poems in Ø1. He studied Mongolian1/*something exotic1 to speak it1 to his mother-in-law. He studied Mongolian1/*something exotic1 to speak Ø1 to his mother-in-law. She set aside a week1/some period of time1/*something1 to spend it1 with her fiancé. She set aside a week1/some period of time1/*something 1to spend Ø1 with her fiancé.

Note that the starred form of (56c) is irrelevantly grammatical on a reading where something denotes a location and not an NL, in which case the controlled position also then denotes a location. Third, the chromaticity dimension is relevant to the sort of marginal ‘instruction set’ nonsubject control touched on in Postal (1994b): (57)

a. Choose some comfortable position1/*something1 and assume it1 for a short period. b. Choose some comfortable position1/*something1 comfortable and assume Ø1 for a short period.

Fourth, chromaticity status is relevant to the nonsubject control characteristic, according to Postal (1993a, 1994b, 2001a, 2001b), of parasitic gaps (but see Levine, Hukari, and Calcagno, 2001, for a strongly opposing opinion): (58)

a. What language1/*What1 did they discuss/argue about t1 immediately after hearing the spy speak it1? b. What language1/*What1 did they discuss/argue about t1 immediately after hearing the spy speak Ø1?

(59)

a. That is a holiday1/*something1 that Freddy thought fondly of t1 without ever spending it1 in Naples. b. That is a holiday1/*something1 that Freddy thought fondly of t1 without ever spending Ø1 in Naples.

3.5. Chromatic agreement and restrictive relative clauses Just as restrictive relative pronouns, visible or not, appear to agree with the heads of the nominals they form in gender and number, as illustrated in (60), they can be argued to show agreement along the chromaticity dimension as well: (60)

a. the woman who praised herself/*himself/*themselves b. the man who praised himself/*herself/*themselves c. the children who praised themselves/*herself/*himself

CHROMATICITY

151

First, consider cases where the overall DP formed with a restrictive relative is itself chromatic: (61)

a. Some language/*Something that Teresa spoke Ø was discussed at length. b. Some position/*Something that the gorilla assumed Ø was discussed at length.

Evidently, when the overall DP is chromatic, its contained relative clause can contain a relative extraction gap in a position that is required to be chromatic, but this is impossible if the overall DP is nonchromatic. This would follow if the relative pronoun (visible or not) had to agree with the head DP in chromaticity value. The parallel point is made when one chooses relative clause gaps in positions that are required to be nonchromatic. This is only possible when the head DP is also nonchromatic: (62)

a. The thing/*problem that Ø is eating Gilbert is unknown. b. Something strange/*Some strange problem that Ø used to be the matter with my car is now affecting Ted’s car as well. c. The thing/*name/that they named her Ø was strange.

All these facts would follow if a principle like (63) held: (63)

A restrictive relative pronoun agrees in chromaticity value with its head DP.

3.6. Chromatic agreement and nonrestrictive relative clauses It should not be surprising at this point to observe that nonrestrictive relative pronouns also manifest chromaticity agreement: (64)

a. Quentin discussed *something/some period of time, which he spent in Bermuda. b. Quentin diagrammed *something/some position, which he was unable to assume . c. The manager was counting on *something/a run, which Tim was unable to drive in .

These nonrestrictive relative contrasts would follow from the independently existing constraints that define chromatic and nonchromatic positions if the following principle holds: (65)

A nonrestrictive relative pronoun agrees in chromaticity value with its head DP.

3.7. Chromatic agreement: conclusion I have argued that strong grounds for the reality of the chromaticity dimension in English grammar are presented by what I have suggested are agreement phenom-

152


ena. These are of two types: first, agreement that involves superficially overt items that include ordinary definite pronouns, restrictive relative pronouns, and nonrestrictive relative pronouns; and second, agreement that involves invisible or covert elements that include controlled pronouns and restrictive relative pronouns. In all cases, modulo the fact that the chromaticity dimension is not morphologically marked, the facts of chromaticity value agreement seem parallel to those for person, gender, and number agreement, which argues that chromaticity is a grammatical dimension in the same sense that these traditional categories are.10, 11

4. Ellipsis phenomena I believe the chromaticity dimension also has implications for the description of ellipsis phenomena like Gapping, Nominal Gapping, Pseudogapping, VP Deletion, and Comparative Deletion. Consider: (66)

Gapping (see, e.g., Jackendoff, 1971; Lobeck, 1995; and Nejit, 1980) a. Jerome spoke some African language and Carol some Asian one. b. Jerome discussed something African and Carol something Asian. c. Jerome spoke some African language and Carol some Eurasian language/*something Eurasian. d. Something simple is the matter with my car and something complex with Ted’s. e. *Something simple is the matter with my car and some very bad problem with Ted’s.

(67)

Nominal Gapping (see Jackendoff, 1971) a. Lucille’s spending of a week in Bermuda and Janet’s of a month in Spain were unusual. b. *Lucille’s spending of a week in Bermuda and Janet’s of something even more indulgent in Spain were unusual. c. Lucille’s assumption of a common position and Janet’s of a hitherto unknown position were unexpected. d. *Lucille’s assumption of a common position and Janet’s of something hitherto unknown were unexpected.

(68)

Pseudogapping (see Baltin, 2003; Lasnik, 1999b; and Levin, 1978, 1986) a. Although Jerome did not speak any African language, he did some Asian one. b. *Although Jerome did not speak any African language, he did something Asian. c. Although nothing is the matter with my car, something serious is with Ted’s. d. *Although nothing is the matter with my car, some serious problem is with Ted’s.

(69)

VP Deletion (see, e.g., Fiengo and May, 1994; Johnson, 2001; and Sag, 1976) a. Some African language Terry did speak, but some Eurasian language/*something well known in America he didn’t. b. Although nothing seemed to be the matter with my car, something/*some problem/ *a serious problem was.

CHROMATICITY

(70)

153

Comparative Deletion (see Bresnan, 1973, 1975; and Kennedy, 1997a) a. Susan has studied the same language that Ted discussed/speaks. b. Susan has studied the same thing that Ted discussed/*speaks. c. Susan diagrammed a different position than Ted discussed/assumed . d. Susan diagrammed a different thing/something different than Ted discussed/ *assumed .

These data show that the same selectional restrictions found in clauses with visible selectors like speak, the matter with, spending, assumption, and so forth show up in clauses (and nominals) that manifest ellipsis of various sorts determined by phrases that contain those restricted selectors. This supports a view that this sort of ellipsis actually involves invisible versions of the syntactic entities that determine the selections. It also suggests that the zeroed constituents contain actual DPs of the sort that can manifest the chromatic or nonchromatic categorization.

5. Types of nonchromatic DPs It was seen in section 1 that one type of nonchromatic DP required a single-word combination of D + noun, as argued by the positioning of simplex adjectives after the nonchromatic noun: (71)

a. b. c. d.

Something terrible took place. Some terrible event took place. They did not witness anything terrible. They did not witness any terrible event.

Now, it was claimed but not explicated that the one-word property, analyzed by Kishimoto (2000) in terms of N-raising, is only characteristic of one subtype of nonchromatic DP. Given what has been established about the existence of certain selectional positions that only allow nonchromatic DPs, one can support the earlier claim by showing that in such positions one finds both one-word and multiword DPs, hence multiword nonchromatic DPs. Consider then: (72)

a. b. c. d. e. f. g. h.

(Only) Something minor was the matter with her liver. *Some problem/*Some disease/*Some rare condition was the matter with her liver. A terrible thing was the matter with her liver. The awful thing that was the matter with her liver caused her to be hospitalized. The only thing that was the matter with her liver was not serious. That frightening thing was also the matter with my liver. Certain frightening things might be the matter with his liver as well.12 Herb doesn’t believe a damn thing to be the matter with her liver.

Cases like (72b) argue, as already indicated earlier, that the subject of the matter with must be a nonchromatic DP. But then cases like (72c, d, e, f, g, and h) show clearly that some nonchromatic DPs do not manifest the D + N as a single-word feature.

154


I do not have a good understanding of the DP internal structural factors that distinguish the distinct types of nonchromatic DPs, but roughly it may work as follows: First, the single-word feature is clearly not associated with definite DPs, as in (72e, f). Second, it seems that there is a division among nondefinite DPs that cuts along the lines that separate some from a forms. So, informally: (73)

The single-word feature (N-raising in Kishimoto [2000]’s terms) is associated only with indefinite DPs of the some type.

An obvious observation is that for (73) to have any chance of being viable, it is necessary to analyze a range of Ds that do not manifest surface instances of the morpheme some as being of the some type. These include most obviously anything, nothing, everything, what, and whatever, as in (74): (74)

a. b. c. d. e. f.

She doesn’t believe anything serious to be the matter with the liver. She believes nothing serious to be the matter with the liver. (Just about) Everything conceivable is the matter with the liver. What does she believe to be the matter with the liver? No matter what she believes to be the matter with the liver, . . . Whatever she believes to be the matter with the liver is likely to be serious.

Less obviously, there is also a range of slang forms that include those of (75) and need to be categorized as indefinite some forms:13 (75)

a. Harry proved stuff to be the matter with the liver. b. Harry proved zilch/zip/zippo/zero to be the matter with the liver. c. Harry did not prove beans/crap/dick/diddley/diddley-squat/fuck-all/jack/jack-shit/ jack-squat/piss-all/poo/shit/shit-all/squat to be the matter with the liver.

These conclusions seem to me to have no known undesirable consequences.

6. Coordination of chromatic and nonchromatic DPs When DPs of distinct categories coordinate, the resulting complex DP is assigned to a category as a function of the categories of the conjuncts. This is most visible in agreement phenomena. So, in an NL with both dual and plural number, the conjunctive coordination of two singulars yields a dual and the coordination of a singular and dual or plural yields a plural.14 In English, with no dual/plural distinction, conjunctive coordination of any DPs yields a plural: (76)

a. b. c. d.

Mike is tall. Roberta is tall. Mike and Roberta are tall. Mike and those two runners are tall.

CHROMATICITY

155

With respect to person, in English and possibly universally coordination of a first person with any person yields a first person (plural) and coordination of a second person DP with any non–first person DP yields a second person DP: (77)

a. b. c. d. e.

We are proud of ourselves. You are proud of yourselves. You and I are proud of ourselves/*yourselves. Those guys and I are proud of ourselves/*themselves. You and those guys are proud of yourselves/*themselves.

Strangely, though, it seems that conjunctive coordination of chromatic and nonchromatic DPs yields a complex DP that cannot be taken to be either exclusively chromatic or exclusively nonchromatic. The evidence is that apparently no mixed conjunctive coordination of chromatic and nonchromatic DPs can occur in any context restricted to either chromatic or nonchromatic forms. This is true both for pure selectional cases and for those that involve what I have taken to be pronominal agreement: (78)

Required Chromatic Context a. Renee attended <matriculation> some school/*something expensive. b. *Renee attended <matriculation> at various times some private school and something inexpensive. c. *Renee described some private school1 and something cheap2 after attending <matriculation> them1,2.

(79)

Required Nonchromatic Context a. Something grave/*Some minor problem was eating Gilbert. b. *At different times something grave and some minor problem were eating Gilbert. c. *Renee described something grave1 and some minor problem2 after learning they1,2 were eating Gilbert.

It is unclear to me how one should react to this data. One possibility would involve a claim that such coordinate phrases must be taken to be both chromatic and nonchromatic, which indicates that one cannot analyze these in a way equivalent to taking them to be plus and minus values of a single binary feature. Another possibility would be to claim that such coordinations are assigned to neither the chromatic nor nonchromatic categories. Not having studied these facts in any detail, I am not in a position to offer anything that chooses between the alternatives or to offer any distinct solution.

7. A prescriptively disfavored type of agreement Normally, English morphologically singular DPs determine, not surprisingly, morphologically singular pronominal agreement:

156 (80)


a. The doctor1/That doctor1/Some doctor1/Every doctor1 claimed he1 was ethical. b. A professor1/This professor1/A certain professor1/No professor1 was sure she1 was right.

Replacing he/she by they in these cases would preclude the antecedent relation that is marked. However, there are cases where plural pronominal agreement that involves they, them, or their is possible, although prescriptively looked down upon: (81)

a. b. c. d. e.

If everyone1/someone1/no one1 thinks they1 are a genius/their1 child is a genius, . . . Who1 here thinks they1 are a genius? No matter who1 thinks demons are after them1, . . . Whoever1 thinks they1 are/their1 child is a genius is probably a moron. Anyone who1 thinks they1 are a genius is probably a moron.

While the regular, expected singular agreement is also possible in such cases, apparently only the human subclass of the same subset of nonchromatic forms that permit else permits this “fake” plural agreement: (82)

a. *If a philospher1/a certain guy1/a specific fellow1 thinks they1 are a genius/their1 child is a genius, b. *What guy1 thinks they1 are a genius? c. *No matter what guy1 thinks demons are after them1, . . .

Those are at least my judgments. But the “Oddments and Miscellanea” column of the Vocabulary Review 3, no. 7 (whose prescriptive motto is “A society is generally as lax as its language”), of July 2001, states: Each month, “Oddments and Miscellanea” will focus on a particular matter of faulty grammar, slipshod syntax, or improper punctuation. This month’s admonition: Avoid using the plural pronoun their, them, or they following words like each and one, every and any, everyone and everybody, anyone and anybody, someone and somebody, and no one and nobody when the antecedent is clearly singular.

The prescribers then present the following list of putatively bad examples and their proposed improvements: (83)

a. Each of the women during this eight-week program developed their body as well as their mind and emotions. USE her; her. b. Everyone has their own story. USE his or her or his or her. c. No one wants their name and information given to anyone and we at Elante Luggage hold this to be paramount to good business. USE his or her or his or her. d. A quick e-mail to thank somebody for their time goes a long way. USE his or her or his or her. e. Every international student when they first came to Toorak College must have felt a bit nervous and homesick. USE he or she or he or she. f. If you want to see what someone truly feels they deserve, just take a look at what

CHROMATICITY

157

they have. USE he deserves or she deserves or he or she deserves; he has or she has or he or she has. g. Every one of the contestants wore a patch on their vests. USE his vest or her vest or his or her vest. h. How do you tell someone that you love them? USE him or her or him or her. i. It’s time for anyone who still thinks that singular “their” is so-called bad grammar to get rid of their prejudices and pedantry! USE his or her or his or her.

As prescriptive grammar goes, this is high-quality. The authors provide a rich list of examples and a clear indication of the construction they think should be avoided. Note, though, that while their preexemplification text cites essentially nonchromatic antecedents, of their nine examples only six involve nonchromatic antecedents. For me, such a heterogeneous collection misses a distinction. That is, I regard (83a, e, and g) as truly ungrammatical, while the other six examples are fine. It is unclear that the Vocabulary Review’s prescriptive remarks are incompatible with my claim that plural agreement with singulars is only possible with a (subtype) of nonchromatic DP antecedent. Because, arguably, for its authors, all such agreement is ungrammatical. What is unclear is whether the sentences they believe they are proscribing, including (83a, e, and g), really occur and “need” proscribing or whether they have just been made up randomly to illustrate a point by authors concerned only to devalue all plural agreement with singular antecedents. The evidence in Pinker (1994: 390–391) is also a bit equivocal. Six of the seven examples of ‘fake agreement’ that he cites have nonchromatic antecedents. The seventh, said to be a quote from J. D. Salinger’s Catcher in the Rye, was: (84)

He’s one of those guys who’s always patting themself on the back.15

But I find this ungrammatical, which raises the issue of dialect variation. In any event, my view that the existence of the plural agreement with singular antecedence phenomenon supports the relevance to English grammar of the chromatic/nonchromatic distinction is supported if the following holds. There are some speakers like I claim to be who restrict such antecedence to that subtype of nonchromatic antecedents that determine one-word character and sanction a following else.16

8. Limitations of informal statements It will not have escaped the reader that the discussion of this chapter has been entirely informal. I have not provided a precise account of the internal structure of DPs to support the claim that English DPs divide into the chromatic and nonchromatic. Nor have I spelled out mechanisms that could account for the conclusion, argued in some detail, that chromaticity is a dimension relevant for agreement of various types. These are serious limitations but, fortunately, not ones directly relevant to the chief point of this essay, which is merely to document the existence of the chromatic/ nonchromatic distinction, one that past linguistics seems to have overlooked.

158


One point, though, that is worth expanding is this: Just as, for example, nouns like man/woman, boy/girl, and such must be the core locus of the distinction of grammatical gender, just so nouns must be the core locus of the distinction between chromatic and nonchromatic. Since, however, the distinction ultimately divides large DP constituents into corresponding types, it is evident that a proper grammatical framework must allow for some kind of ‘projection’ of the lexical noun distinction to one that distinguishes as well their minimal containing DPs. There are various extant approaches to such questions and I have nothing to add to the understanding of the matter here.

5

The Structure of One Type of American English Vulgar Minimizer

1. Background In the summer of 1995, Haj Ross and I began jointly looking into the grammar of the vulgar American English slang forms in (1): (1)

SQUAT = Vulgar Minimizers = {beans, crap, dick, diddley, diddley-poo, diddley-squat, fuck-all, jack, jack-shit, jack-squat, piss-all, poo, shit, shit-all, squat}

Some examples of some uses of these forms, taken to be members of the nonchromatic class in chapter 4, are given in (2): (2)

a. b. c. d.

Olmstead didn’t say dick about the new dean. Olmstead knows fuck-all about Botswana. Olmstead didn’t contribute jack to the emergency fund. Olmstead doesn’t understand squat about topology.

In naming these expressions as I have, I intend to bring out what I take to be a genuine connection with other minimizers, a standard category of negation studies; see Horn (2001: 452–453). Generally, a minimizer is a type of DP that denotes minimal elements on some scale: (3)

Minimizers that are not vulgar minimizers include, for example, a drop/a word/a red cent/a finger, as in (4): 159

160 (4)


a. b. c. d.

Sally didn’t drink a drop. (minimal amount of liquid) Stan didn’t say a word. (minimal amount of linguistic utterance) Sarah didn’t have a red cent. (minimal amount of money) Steve didn’t lift a finger. (minimal amount of effort)

One key difference between vulgar minimizers and others such as those in (4) is that mostly the former are not narrowly restricted to particular dimensions but can express minimality along many dimensions not specifically invoking animate things, because, like most minimizers, all those of (1) form exclusively inanimate nominals. A few English minimizers, specifically those highlighted in (5), do not: (5)

a. Spiro didn’t see a soul on Sunday. (minimal number of people) b. Spiro didn’t see a living soul on Sunday. (minimal number of people) c. The police didn’t find a fucking soul alive in the crack house. (minimal number of people)

In discussing a linguistic subject, one normally references the literature on it, which is easy in this case, given that, as far as I know, grammatical/semantic works devoted to or containing detailed analyses of English vulgar minimizers are nearly nonexistant. See, though, Horn (1996b, 2001) for some discussion and Nexis search examples. While (1) lists many forms, no doubt most Americans use or even know only a proper subset. I assume, though, that most American speakers are familiar with at least one member of SQUAT and thus have introspective access to the phenomenon.1 Various vulgar minimizers are mentioned in slang dictionaries. There they appear to uniformly be said to mean “nothing,” as are other forms not listed in (1) like those in (6): (6)

Z-minimizers = {zero, zilch, zip, zippo}

See, for instance, Spears (1996: 374), which mentions squat and diddly-squat and gives only the meaning “nothing.” However, something key is missed by such dictionaries. As already in effect illustrated in (1), while vulgar minimizers do have uses in which they seem to mean “nothing” and in this respect are like the forms of (6), many of them for many speakers also have different uses in which they function as negative polarity items (NPIs), that is, items that cannot occur in simple non-negative clauses but do occur with negation. This polarity property is illustrated for the animate minimizer in (5c) in (7): (7)

a. b. c. d.

Claudia did not see a fucking soul. *Claudia saw a fucking soul. Claudia did so see someone. *Claudia did so see a fucking soul.

Notably, none of the Z-minimizers occurs as an NPI in the NPI licensing environment of (7a), as (8) illustrates:

THE STRUCTURE OF ONE TYPE OF AMERICAN ENGLISH VULGAR MINIMIZER

(8)

161

??Claudia did not see zilch/zip/zippo.

If (8) is acceptable at all, it can only be as a non-NPI instance of denial negation, with strong stress on not, representing rejection of a previous claim that Claudia saw zilch. Seemingly, even in an NPI-accepting environment, the Z-minimizers can only have the non-NPI meaning “nothing.” But for vulgar minimizers, the situation is different, as illustrated in (9) and (10): (9)

(10)

a. Claudia saw squat. b. Claudia did not see squat. a. Claudia discovered dick. b. Claudia did not discover dick.

While the a. forms are consistent with the behavior of Z-minimizers and with the slang dictionary claim that vulgar minimizers mean “nothing,” the b. cases seem to indicate that they can also mean “anything.”2 In fact, although Spears (1996) gives only “nothing” as the meaning of squat, one of his two example sentences is of the type in (9b), where squat would naturally be taken to mean “anything” rather than “nothing.” (11)

Spears (1996: 374) I worked all day on this, and she didn’t pay me squat.

Besides raising a problem for slang dictionary meaning claims, pairs like (9) and (10) are truly extraordinary in one clear respect, for they illustrate forms that occur in a fixed position where the presence versus absence of an overt negation seems to make no semantic difference. That is, (9a, b) and (10a, b) are logical (truth functional) equivalents. There are few other such forms.3 The standard and expected situation where presence of a negative yields distinct meanings is seen in (12): (12)

a. b. c. d.

Claudia Claudia Claudia Claudia

discovered many treasures. ≠ did not discover many treasures. violated every rule. ≠ did not violate every rule.

Focus then on a logically equivalent pair like (13a and c): (13)

a. b. c. d.

Irma Irma Irma Irma

understands dick about clones. = understands nothing about clones. does not understand dick about clones. = does not understand anything about clones.

Given the apparent perfect equivalences here, it is overwhelmingly tempting to attempt to reduce the patterns in the vulgar minimizer cases to that of the standard cases

162


with anything and nothing. The obvious way to do this would be as follows: First, assume that each of the vulgar minimizer pieces of morphology represents a noun stem. Second, claim that the vulgar minimizers are ambiguously analyzable in exactly the same way as either nothing or anything. Assuming that no forms like nothing involve a negative determiner that consists of a syntactic negative (NEG) + some, this means taking one term of the ambiguity for vulgar minimizers to involve something like (14), the other something like (15): (14)

a. nothing = [DP [D NEG + some] + [N thing]] b. squat/dick/ . . . = [DP [D NEG +some] + [N squat/dick/ . . .]]

(15)

a. anything = [DP [D any] + [N thing]]4 b. squat/dick/ . . . = [DP [D any] + [N squat/dick/ . . .]]

The idea for (14b) and (15b) would then be that the determiners that otherwise manifest as no and any are obligatorily invisible when they occur with a vulgar minimizer noun. Let me distinguish terminologically two forms of vulgar minimizers: those of (9a) and (10a) I will refer to for reasons that will emerge as type Z, and those of (9b) and (10b) I will refer to as type A, this intended to suggest a link with anything. The central goal of this chapter is to argue for the following (at least to me) quite amazing conclusion: Any equation like that in (14b), that is, any analysis that treats the nonpolarity variant of each member of SQUAT as in effect a negative form like nothing, is completely wrong. Z-type vulgar minimizers cannot be analyzed as forms having negative quantifiers for determiners and, even more generally, cannot be correctly analyzed as containing any syntactic negatives at all.

2. What type Z vulgar minimizers are not 2.1. Remarks Thus the goal of this section is to show that in sentences like (13a) the vulgar minimizer dick does not contain a negative quantifier and, more generally, that such sentences contain no syntactic negation whatever. Admittedly, it is unclear that anyone has publicly claimed that they do contain such. The few remarks in the literature about such forms seem quite equivocal about their analysis: (16)

van der Wouden (1997: 78) “For instance, one of the many possibilities to affectively strengthen negation doesn’t use negation at all.”

The author then cites the type Z vulgar minimizer case (17a) and notes that “[a] variant of this sentence does use negation,” citing (17b): (17)

a. He knows shit about GB. b. He doesn’t know shit about GB.


163

One might take these remarks to deny that (17a) involves negation, but the conclusion is hardly solid. Similarly: (18)

Hoeksema, Rullman, Sánchez-Valencia, and van der Wouden (2001: viii–ix) “The paper includes a discussion of squat and similar expressions of scatological origin, which have recently undergone part of the Jespersen Cycle, rendering them negative in force even in the absence of any overt negation: They told him squat = They didn’t tell him squat.”

Again it is unclear what inference to draw about the authors’ view of the role of syntactic negation in the analysis of the first, type Z, example. Finally: (19)

Horn (2001: 185) “The key fact here, as recognized by the OED, is that squat can appear either in the scope of a licensing negation as the equivalent of anything, or on its own as the counterpart of nothing.”

Horn refers to the former usage as licensed squat, the latter as unlicensed. Once more, though, it is not evident whether the remarks should be read as a claim that the correct synchronic analysis of type Z vulgar minimizers involves negation or not. But regardless of whether anyone else has assumed type Z vulgar minimizers involve a syntactic negative, the position to be criticized here is not a straw man, since I myself adopted it as a working hypothesis for many months, struggling vainly to keep the view consistent with the collection of contrary data that continued to accumulate. I will try to indicate why such consistency is not possible by documenting a variety of ways in which type Z vulgar minimizers behave as if syntactic negation was entirely absent. It turns out, I believe, that only two indications even suggest that type Z vulgar minimizers involve negation. The first is simply the morphological parallelisms between type Z and type A minimizers; the second is the logical equivalence between pairs like (9a, b) and (10a, b). A logic that leads from these facts to a view that type Z vulgar minimizers do contain negation might go something like this: Relevant pairs like (9a, b) and (10a, b) are clearly related morphologically and yield logical equivalents. Given that the type A variants do involve syntactic negation, which is easily shown, then assuming that the logical equivalent of something negative is (double negations aside) something negative, the conclusion is reached. It should be of some general interest that this pattern of reasoning fails, as I will now argue. 2.2. Double negatives Despite what one heard in school about the evils of double negatives, even standard American English allows double negative cases like (20a), to be sharply distinguished from exclusively substandard cases like (20b); see Postal (in press). (20)

a. No professor favors NO proposal/NOTHING. (caps = contrastive stress) b. substandard: No professor favored no proposal/nothin’. = ‘No professor favored any proposal/anything’

164


The substandard (20b) involves so-called negative concord. But the standard (20a) involves something else and is equivalent to (21): (21)

Every professor favors SOME proposal.

See also the fine (22a), which contrast with the only substandard (22b): (22)

a. The professor did not favor NO proposal = ‘favored SOME proposal’. b. substandard: The professor did not favor no proposal = ‘did not favor any’.

Now, since standard English no forms like nothing permit non-negative concord structures like (20a) and (22a), if Z type vulgar minimizers were disguised no forms, they also would presumably permit that; but they do not, as (23a and b) indicate: (23)

a. *No professor favored SQUAT. b. *The professor did not favor JACK.

I believe such examples are ill formed; but in any event, there is no way they can be interpreted like parallel cases with stressed NOTHING. That is, type Z vulgar minimizers do not show the double negative behavior of standard no forms. 2.3. Positive versus negative interrogative tags A well-known test for the presence of negation first introduced in Klima (1964: 263– 265) involves the possibility of a positive (negation-free) interrogative tag. Notably though, in this respect type Z vulgar minimizer structures behave like positive forms and contrast with both no forms and type A vulgar minimizer structures: (24)

a. b. c. d.

Janet Janet Janet Janet

read some book, *did/didn’t she? read no book, did/*didn’t she? didn’t read squat, did/*didn’t she? read squat *did/didn’t she?

2.4. Emphatics: too/so versus (n)either A second known test for negation also found first in Klima (1964: 261–262) involves the possibility of emphatics of the form (n)either, as opposed to positive emphatics of the form too/so. And with respect to this test, too, type Z vulgar minimizer cases behave like positive forms and not like clear negative forms, which include no forms and type A vulgar minimizer expressions: (25)

a. b. c. d.

Janet read some book and Hilda read some book, too/*either. Janet read no book and Hilda read no book, *too/either. Janet didn’t read squat and Hilda didn’t read squat, *too/either. Janet read squat and Hilda read squat, too/*either.


e. f. g. h.

165

Janet read some book and so/*neither did Hilda. Janet read no book and *so/neither did Hilda. Janet didn’t read squat and *so/neither did Hilda. Janet read squat and so/*neither did Hilda.

2.5. The not even + X strengthener A third known test for negation also introduced in Klima (1964: 262–263) involves the possibility of the strengthening expression not even + X, which requires inter alia a syntactically negative ‘antecedent’. Again type Z vulgar minimizers contrast with type A ones and manifest the behavior of non-negatives: (26)

a. b. c. d.

Jane read some book yesterday (*, not even the assigned book). Jane read no book yesterday (, not even the assigned book). Jane didn’t read squat yesterday (*, not even the assigned book).5 Jane read squat yesterday (*, not even the assigned book).

2.6. I don’t believe/think parentheticals A fourth test for negation involves parenthetical clauses with negation, which only can modify negation-containing clauses. And once more, type Z vulgar minimizer structures contrast with those that contain type A vulgar minimizers or uncontroversial negation-containing expressions: (27)

a. b. c. d.

Jane read some book yesterday (*, I don’t think). Jane read no book yesterday (, I don’t think). Jane didn’t read squat yesterday (, I don’t think). Jane read squat yesterday (*, I don’t think).

2.7. ‘Expression of agreement’ clauses: So versus not A fifth, perhaps less known test for negation, suggested to me by Haj Ross, involves elided clauses that express agreement with the content of a preceding clause. Such clauses manifest an initial yes + so if the clause ‘agreed with’ is positive but an initial no and final not if it is negative. Again, type Z vulgar minimizer structures manifest positive behavior: (28)

a. b. c. d.

Jane read some book yesterday. Yes, I guess so/*No, I guess not. Jane read no book yesterday. *Yes, I guess so/No, I guess not. Jane didn’t read squat yesterday. *Yes, I guess so/No, I guess not. Jane read squat yesterday. Yes, I guess so./*No, I guess not.

2.8. Negative polarity licensing Finally, as the now common jargon has it, no forms can license NPIs, a few of which are highlighted in (29):

166 (29)


a. Helga gave no magazines to anyone. b. Helga has said nothing to me in years. c. Helga learned nothing in any convent at all.

If then, Z type vulgar minimizers were no forms, they would be expected to also license NPIs. But the contrast between (29) and (30) shows they do not:6 (30)

a. *Helga gave squat to anyone. b. *Helga has said squat to me in years. c. *Helga learned squat in any convent at all.

2.9. Summary Overall then, the evidence seems quite consistent and compelling. With respect to a range of fact types that independently differentiate negative from positive clauses, type Z vulgar minimizer forms fail to reveal any syntactic indication of involving negation and their containing clauses behave rigorously like positive clauses. Given their relation to type A minimizer forms, logical equivalences, and such, this might seem to yield a bit of a paradox. But the next section argues that it does not.

3. What type Z vulgar minimizers are Taking it now as justified that type Z vulgar minimizers are not analyzable in the same way as nothing and do not even contain negatives, let us return to logically equivalent pairs like (31): (31)

a. They don’t understand squat. b. They understand squat.

While a priori it could have well seemed plausible to make sense of such pairs by taking type Z vulgar minimizers to be like nothing and type A ones to be like anything, it has been shown that the former assumption is untenable. How then should one analyze type Z vulgar minimizers? My suggestion, stimulated by an idea in Déprez (1997) about certain French forms, is that they should be analyzed as containing a determiner that is not an invisible negation + some but rather an invisible cardinal numeral zero, as in (32):7 (32)

type Z squat = [DP [D zero] + [N squat ]]

This analysis motivates the terminology ‘type Z vulgar minimizers’ that has been adopted. Analysis (32) permits capturing two crucial aspects of type Z vulgar minimizers. First, without any grammatically present negation, they end up, as required by the facts, being logically equivalent to nothing, given the independent existence of truth functional equivalences like (33):


(33)

a. Joanne read no books

167

↔ b. Joanne read zero books.

Second, the zero determiner analysis of type Z vulgar minimizers accounts for the fact that they behave grammatically like they contain no syntactic negatives, since they simply don’t. More generally, type Z vulgar minimizer structures are seen to essentially share most of the properties of independently occurring DPs with the zero determiner, as (34) shows: (34)

a. No woman drank *SQUAT/ZERO martinis. b. Penelope drank zero martinis/squat, did/*didn’t she? c. Penelope drank zero martinis/squat and Francine drank zero martinis/ squat, too/ *neither. d. Penelope drank zero martinis/squat and so/*neither did Francine. e. *Penelope drank zero martinis/squat yesterday, not even weak martinis/ones. f. *Penelope drank zero martinis/squat yesterday, I don’t think. g. Penelope drank zero martinis/squat yesterday. Yes, I guess so. h. *Penelope drank zero martinis/squat yesterday. No, I guess not.

The only real contrasts for my idiolect are that in (34a) and that, while as seen earlier, type Z vulgar minimizers cannot license NPIs, zero DPs can: (35)

a. *Hector sent squat to any of his ex-wives. b. Hector sent zero presents to any of his ex-wives.

But as suggested in note 6, some speakers accept cases like (35a), so this may well be an idiosyncrasy of marginal significance. The difference in (34a) shows that type Z vulgar minimizers are more restricted than explicit zero forms with respect to such contrast structures. Since I do not understand such expressions very well, I will not try to say anything more about them.

4. McCawley’s puzzle 4.1. The puzzle McCawley (1998: 607) presented some observations about the distribution of the word nothing with respect to certain sentence properties often, as in section 2, taken since Klima’s (1964) work to provide tests for the presence of negation. Specifically, McCawley discussed the contrast between (n)either and so/too, noting, as I did in section 2, that the former is only possible if the clause that actually contains it is negative in a relevant sense. McCawley’s currently relevant observation is that, unexpectedly, with nothing as an object (see later discussion for clarifications) but not with no forms in general, it is possible to have either too or either, apparently without meaning difference: (36)

McCawley (1998: 607) a. John said nothing, and Mary said nothing, too. b. John said nothing and Mary said nothing, either.

168


In general, these forms are, of course, not both possible in the same context: (37)

a. Jane cooked no reptile meat and Sheila cooked no reptile meat, either/*too. b. Ferdinand chased neither student and Otto chased neither student, either/*too.

Why (36a and b) are both possible is then quite mysterious and McCawley, noting the problematic character of (36a) for his assumptions, simply said: “I will leave this problem unresolved.” The distinction in (36) is not isolated but illustrates a systematic possibility for the form nothing. Note initially: (38)

a. John contributed *no idea/nothing and so did Mary. (See Jackendoff, 1972: 364.) b. John contributed no idea/nothing and neither did Mary.

4.2. A novel proposal The analysis of type Z vulgar minimizers proposed in section 3 offers a new and intriguing way to look at the relevant exceptional alternation. For there is evidently an abstract similarity between the data of section 4.1 and that already encountered. In the case of type Z vulgar minimizers, one finds forms that in some vague sense seem to be negative or equivalent to negatives like nothing but behave according to syntactic tests for negation like non-negatives. Arguably, that is just what is found in (36a) and (38a). I suggest then that the two versions of (36) and (38) differ in that the one that can occur with either contains a negation, specifically an ordinary no form, while the one that can occur with too fails to contain a negation. What is it then? The answer that previous sections suggest at this point is that it has the structure of a type Z vulgar minimizer. This means that one could allow for a certain case in which the determiner zero + some vulgar minimizer stem is pronounced nothing, which then might have at least the two distinct analyses in (39): (39)

a. nothing = [DP [D NEG + some] + [N thing ]] b. nothing = [DP [D zero] + [N, vulgar minimizer thing ]]

Required then would be that at least one usage of thing be a member of the class of regular noun stems and also a member of the restricted class of vulgar minimizer nouns, one of which only takes the zero determiner.8 A problem with (39b), though, is that it requires taking the vulgar minimizer stem that underlies the positive nothing to be an exception to the general rule that the D associated with a vulgar minimizer is phonologically null. And it requires ad hoc morphology that realizies the determiner zero as no. A superior alternative would then be to retain the view that that the null realization rule for Ds with vulgar minimizer nouns is absolute and to analyze the nonnegative nothing instead slightly differently as in (40): (40)

nothing = [DP [D zero] + [N, vulgar minimizer nothing ]]


169

Analysis (40) has the advantage over (39b) of not complicating the general rule for D invisibility with vulgar minimizer nouns. And it also requires no special irregular realization of the D as no. Beyond the fact that the proposal to analyze the positive-behaving nothing as a type of vulgar minimizer DP immediately reduces McCawley’s (1998) observation (36a) to regularity, it is strongly supported by a variety of other evidence, all of which shows that on one analysis nothing in positions like that of (36) behaves like a negative form and on another like a positive one. 4.3. Evidence for treating some instances of nothing as type Z vulgar minimizers First, recall that at least in my idiolect type Z vulgar minimizers cannot license NPIs. If then the nothing that goes with too/so is not negative and has the structure of a type Z vulgar minimizer, while that which goes with (n)either is negative, one would expect that NPI licensing is possible for speakers like me only with the latter, which is correct: (41)

a. b. c. d. e.

Claudia said nothing/*squat at any time and I said nothing at any time, either. *Claudia said nothing at any time and I said nothing at any time, too. Claudia said nothing at all and I said nothing at all, either. *Claudia said nothing at all and I said nothing at all, too. Claudia said nothing to anyone and *so/neither did I.

Second, while no forms can freely be the targets of so-called negative fronting, type Z vulgar minimizers cannot be. Notably, when nothing is a negative fronting target, only behavior consistent with the negative analysis is possible: (42)

Nothing/*Squat did Claudia say to Henry and nothing did Louise say, either/*too.

Third, as seen earlier, type Z vulgar minimizers do not, in contrast to negative forms, permit positive tags in questions but, again in contrast to negative forms, do permit negative tags. Notably, tags don’t combine freely with too/either; when either is present, only the tag appropriate for a negative clause is possible: (43)

a. Jerome ate nothing and Stan ate nothing, either, *didn’t/did he? b. Jerome ate nothing and Stan ate nothing, too, didn’t/*did he?

Fourth, no forms but not type Z vulgar minimizers can be strengthened with not even phrases. Strikingly though, in the context here, that is possible only with either, not with too: (44)

The president said nothing and the vice president said nothing, either/*too, not even good-bye.

170


Fifth, no forms but not type Z vulgar minimizers permit negative parentheticals of the form I don’t believe/think. And once more, such are compatible with either but not with too: (45)

a. The president said nothing and the vice president said nothing, either/*too, I don’t think. b. The president said nothing and the vice president said nothing, either/too, I think.

Sixth, no forms but not type Z vulgar minimizers permit the negative version of the ‘agreement’ clause structure: (46)

a. Karen sent no gifts to them, did she? *Yes, I guess so/No, I guess not. b. Karen sent squat to them didn’t she? Yes, I guess so/*No, I guess not.

And once more, in this context, nothing can behave like a type Z vulgar minimizer: (47)

a. Karen sent nothing to them, did she? *Yes, I guess so/No, I guess not. b. Karen sent nothing to them, didn’t she? Yes, I guess so/*No, I guess not.

A different class of arguments can also support the claim that the positive-behaving instances of nothing should be analyzed in the same way as type Z minimizers. These arguments depend on various distributional restrictions on vulgar minimizers that are independent of negation. First, such minimizers cannot appear as passive by phrase complements and indeed not in many types of PPs: (48)

a. Jerome was shocked by nothing/*squat. b. The riot was caused by nothing/*dick. c. That claim is supported by nothing/*jack.

Notably then, when found in these environments, occurrences of nothing show only negative behavior, not the positive behavior associated with type Z vulgar minimizers: Jerome Jerome Jerome Jerome

was shocked by nothing, was/*wasn’t he?9 was shocked by nothing and Irma *was, too/wasn’t, either. was shocked by nothing, (I don’t think) (not even the deception). was shocked by nothing. *Yes, I guess so/No, I guess not.

(49)

a. b. c. d.

(50)

a. Herb sat on top of nothing/*squat. b. Herb leaned against nothing/*jack-shit.

(51)

a. b. c. d.

Herb sat on top of nothing, did/*didn’t he? Herb sat on top of nothing and Jane *did, too/didn’t, either. Herb sat on top of nothing (I don’t think) (not even the table). Herb sat on top of nothing. *Yes, I guess so/No, I guess not.


171

Further, vulgar minimizers cannot occur as the first object of double object constructions: (52)

a. Cynthia gave nothing/*squat that sort of attention. b. Cynthia gave nothing/*jack-shit a glance.

And again, when nothing occurs in this environment, it cannot manifest positive behavior: (53)

a. b. c. d.

Cynthia gave nothing that sort of attention, did/*didn’t she? Cynthia gave nothing that sort of attention and Jane *did, too/didn’t, either. Cynthia gave nothing that sort of attention (I don’t think) (not even her finances). Cynthia gave nothing that sort of attention. *Yes, I guess so/No, I guess not.

In addition, vulgar minimizers do not permit modifying adjectives, relative clauses, or PPs: (54)

a. They sent nothing/*squat useful/ that I liked/of interest. b. They discussed nothing/*jack-shit important/ that I was concerned with/from your book.

And when nothing appears with modifiers, it cannot show positive behavior: (55)

a. b. c. d.

They sent nothing useful/that I liked/of interest, did/*didn’t they? Sally sent nothing useful/that I liked/of interest and Jerome *did, too/didn’t, either. Sally sent nothing useful/that I liked/of interest (I don’t think) (not even a book). Sally sent nothing useful/that I liked/of interest. *Yes, I guess so/No, I guess not.

Next, vulgar minimizers are incompatible with so-called A-adverbs like absolutely, almost, nearly, just about, practically, and virtually; see Horn (2000: 161); all of these combine happily with ordinary no forms: (56)

a. b. c. d.

Harriet saw (*absolutely/*almost/*vitually) squat. Stan regrets (*just about/*practically) jack. Harriet saw (absolutely/almost/vitually) no cormorants. Stan regrets (just about/practically) no decision.

As expected at this point, when nothing is modified by an A-adverb, it can only show negative behavior: (57)

a. Harriet saw absolutely/almost nothing and Ruth saw absolutely/almost nothing, *too/ either. b. Harriet saw absolutely/almost nothing, did/*didn’t she? c. Harriet saw absolutely/almost nothing and *so/neither did Ruth. d. Harriet saw absolutely/almost nothing. *Yes, I guess so/No, I guess not.

172


Other environments for some reason preclude clear vulgar minimizers and when nothing appears in these environments, it must show negative behavior: (58)

a. Ted was near nothing/*squat. b. Ted was fond of nothing/*squat.

(59)

a. b. c. d.

Ted was near nothing, was/*wasn’t he? Ted was near nothing and Irma *was, too/wasn’t, either. Ted was near nothing (I don’t think) (not even a water fountain). Ted was near nothing. *Yes, I guess so/No, I guess not.

(60)

a. b. c. d.

Ted was fond of nothing, was/*wasn’t he? Ted was fond of nothing and Irma *was, too/wasn’t, either. Ted was fond of nothing (I don’t think) (not even poetry). Ted was fond of nothing. *Yes, I guess so/No, I guess not.

It is also the case that for me, type Z vulgar minimizers cannot appear as predicate nominals. And when nothing does, it must show negative behavior: (61)

a. As for that scratch, it is nothing/*squat b. It turned into nothing/*jack-shit.

(62)

a. b. c. d.

That scratch is nothing, is/*isn’t it. The first scratch was nothing and the second *was, too/wasn’t, either The scratch was nothing (I don’t think) (not even an annoyance). The scratch was nothing. *Yes, I guess so/No, I guess not.

Finally, vulgar minimizers seem to be barred from subject positions; and when nothing appears there, it cannot display positive behavior: (63)

a. b. c. d. e.

Nothing/*Squat happened. For nothing/*squat to go wrong would be surprising. Nothing happened, did/*didn’t it? Nothing happened yesterday and nothing happened today, *too/either. Nothing happened yesterday. *Yes, I guess so/No, I guess not.

4.4. Conclusion I hope to have shown in this section that the curious behavior of the form nothing noted by McCawley receives a rather elegant account if it is recognized that in addition to a standard analysis as a DP with a negative quantifier D, nothing has a distinct analysis under which it is in effect a type Z vulgar minimizer based on a noun stem nothing taking the same zero D as other type Z vulgar minimizers. If this account is viable, it shows that the study of vulgar minimizers not only is of interest in itself but also impacts in a positive way the treatment of more standard English forms.

6

The Openness of Natural Languages

1. Curious implications It might seem plausible to the nonspecialist that a given NL, NLx permits one to report linguistic performances, both performances of NLx elements and those of NLs distinct from NLx. By “reporting linguistic performances” I refer to nothing more arcane than forming statements like “Amanda just shouted, “Where’s my baby?’” It might also seem to a nonspecialist that NLx permits one to do descriptive linguistics, not only the descriptive linguistics of NLx but also that of other distinct NLs. By “doing descriptive linguistics” I mean nothing more exotic than forming sentences like “The German word for ‘air force’ is ‘Luftwaffe.’” But while these nonspecialist assumptions might seem not only plausible but also self-evidently true, modern linguistics in its dominant instantiation, generative grammar, in fact denies both these claims. Of course, it does this only implicitly and most advocates of generative grammar may be unaware that its doctrines taken literally preclude what any nonspecialist would assume possible. Readers who do not easily accept this conclusion will find their skepticism addressed in what follows, for a major goal of this study is to justify in detail the claim that generative grammar has the evidently intolerable implications just mentioned.

2. Background Near the beginning of the generative grammar movement in linguistics the following claims were made (all emphases mine: PMP): 173

174 (1)


a. Chomsky (1959: 137) “A language is a collection of sentences of finite length all constructed from a finite alphabet (or, where our concern is limited to syntax, a finite vocabulary) of symbols.” b. Chomsky (1959: 137) “Since any language L in which we are likely to be interested is an infinite set, we can investigate the structure of L only through the study of the finite devices (grammars) which are capable of generating its sentences.” c. Chomsky (1959: 138) “The weakest condition that can significantly be placed on grammars is that F be included in the class of general, unrestricted Turing machines.”

Characterization (1a) clearly was not meant to introduce some arbitrary notion ‘language’ but rather assumed that NLs fell under the conditions given: (2)

Chomsky (1957: 13) “I will consider a language to be a set (finite or infinite) of sentences, each finite in length and constructed out of a finite set of elements. All natural languages in their spoken or written form are languages in this sense, since each natural language has a finite number of phonemes (or letters in its alphabet) and each sentence is representable as a finite sequence of these phonemes (or letters).”

That is, it was in effect claimed in (1a) inter alia: (3)

a. b. c. d.

Every NL sentence is finite in size (therefore, e.g., in length). Every NL sentence is exclusively formed from a fixed, finite vocabulary of symbols. Every NL is an infinite collection. Every NL can be generated (recursively enumerated).

Given (3a, b), it is plausible, although not necessary, that the sentence collections that define NLs form (recursively) enumerable sets,1 and hence (3d) and (1c) might well be true. But they might also be false and (3a, b), interpreted as factual claims about NLs, might also be false. If so, (3d) would be false and hence (1c) would also be false. However, as observed in Langendoen and Postal (1984: 15), Chomsky then gave no arguments or evidence for (3a, b, d) in the 1950s. Nor has he since. This meant that in the mid- to late 1950s, claim (1c) was question begging. Chomsky’s never interrupted and current maintenance of it (see later discussion) is arguably worse. For Langendoen and Postal (1984, 1985) rejected all of these claims, focusing in particular on (3a, d), and offered a purported proof (whose conclusion was named the NL Vastness Theorem) that any NL with productive coordination, that is, apparently every NL, including English, was not only not an enumerable set but also not a set at all. Rather full collections of NL sentences were concluded to be what logicians call proper classes; see, for example, Stoll (1979: 319). And it was argued that this was true because, inter alia, NLs not only have individual sentences that are not finite but also contain, contrary to (3a), infinite-sized sentences of every (nonfinite) cardinality.

THE OPENNESS OF NATURAL LANGUAGES

175

Purported proofs can, obviously, contain flaws. But these results were reviewed in major journals and, while they were hardly welcomed, no reviewer claimed to have found any error in the argument.2 Rather, even the unenthusiastic reviews by and large somewhat grudgingly admitted that the argument seemed sound. Moreover, while eighteen years have passed since the publication of the arguments in question, to my knowledge no one has since even claimed to have shown that they were erroneous. Like any proof, that in Langendoen and Postal (1984, 1985) involved fixed premises and a logic. To reject the conclusion, a rational critic would have to reject at least one of these. But the logic of the proof was almost entirely that of a fundamental result in set theory, Cantor’s Theorem.3 It stretches plausibility, to say the least, to imagine that this aspect of the argument can be undermined. That leaves a rational critic with only two choices: either accept the conclusion or find grounds for rejecting the premise. No one has done the latter and yet the conclusion seems not to be accepted. In particular, although the formulation in Langendoen and Postal (1984) was specifically and explicitly aimed at Chomsky’s various formulations, he has never, to my knowledge, addressed the result.4 Instead, despite a history of speaking of his own work as “rational inquiry” and sometimes disparaging other work in linguistics as lacking this virtue,5 he has throughout the intervening period continued to assume, without justification and without dealing with the counterargument, that nothing precludes taking an NL to involve (in fact, as in [4], to be) a computational procedure that recursively enumerates a set of expressions.6 This view is strongly embedded in his so-called minimalist program, which entirely postdates Langendoen and Postal (1984, 1985). So, for instance: (4)

Chomsky and Lasnik (1995: 14–15) “We may think of the language, then, as a finitely specified generative procedure (function) that enumerates an infinite set of SDs. Each SD, in turn, specifies the full array of phonetic, semantic, and syntactic properties of a particular linguistic expression.”

But of course, unless the argument of Langendoen and Postal (1984, 1985) is refuted, claim (4) is just what a rational person cannot think, since: (5)

Langendoen and Postal (1984: 72): The NL Nonconstructivity Theorem No NL has any constructive (= proof-theoretic, generative, or Turing machine) grammar.

Conclusion (5) is evidently a trivial consequence of the NL Vastness Theorem. So it appears that the argument of Langendoen and Postal (1984, 1985) to the effect that NLs are (i) not recursively enumerable sets and (ii) are not sets at all and hence (iii) have no generative/constructive/Turing machine grammars, although both factually and logically unchallenged, was too arcane and too much in conflict with deeply held a priori beliefs to be taken seriously within contemporary linguistic theory. The arcane aspect was localized in two points: First, the argument depends on granting the existence of nonfinite sentences,7 whereas the unargued prejudice that all NL sentences are finite is very strong in linguistics.8 Second, the argument assumed an axiom about the productivity of coordination, a claim that the class of, for example,

176


declarative sentences is closed under coordination, which was not hitherto recognized. From this closure axiom, appealing to Cantor’s Theorem, the existence of nonfinite sentences can be proven. Now, given these arcane aspects and given the sociological fact that the substantively unchallenged argument has not been accepted, it would be significant if it could be shown independently of these seeming truths that NLs fail to be recursively enumerable sets, hence (must) fail to have constructive grammars, on other, less arcane grounds. I believe this can be done by focusing on claim (3b), largely passed over in Langendoen and Postal (1984). Given that no argument or evidence was presented for this in 1959 or since, it obviously should also be a major point of suspicion. The goal of this chapter is to show that (3b) is false; hence by asserting (1a) without argument the author had then begged and has continued simply to beg the question in a way Langendoen and Postal (1984) mostly ignored of whether NLs could have generative/constructive grammars. By focusing on (3b) it will be possible to present a different type of argument for the claim that no NL has a generative/constructive/proof-theoretic grammar. The attraction of the newer form of argument is that, unlike that of Langendoen and Postal (1984, 1985), it depends neither on recognizing nonfinite sentences nor on accepting the axiom about coordinate closure appealed to there. Moreover, the new argument also does not inherently stand on the conclusion that each NL includes more than ℵo sentences, although it provides independent grounds for the latter conclusion. So the goal of this study is to establish the earlier conclusion on different and possibly less controversial grounds.

3. Fixed, finite lexicons Focus on claim (3b) built into (1), which requires that the sentences of NL NLi are formed exclusively from forms listed (hence listable) in a fixed, finite lexicon of NLi (hereafter: LXi). This doctrine enunciated by its author more than forty years ago and never abandoned by him is maintained in, for example, (6):9 (6)

Chomsky (2000c: 120) “The I-language consists of a computational procedure and a lexicon. The lexicon is a collection of items, each a complex of properties (called features), such as the property bilabial stop or artifact. The computational procedure selects items from the lexicon and forms an expression, a more complex array of such features.”

The author thus makes it clear that the computational procedure he posits as characterizing an NL (or, in his terms, as being an NL) operates by finding items in a lexicon, that is, in some kind of finite list. One can state the doctrine a bit more precisely, if still informally, as in (7a) or (7b): (7)

a. For every NL NLi, every minimal form of every sentence of NLi either is drawn from LXi, or is computable from grammatical mappings of other forms drawn from LXi.


177

b. Let a grammar G be some full computational specification of an NL, NL1 and let F be any arbitrarily chosen minimal form in an arbitrary well-formed sentence of NL1. Then G mentions F.

By ‘minimal form’ I mean roughly what are called morphemes. The notion ‘mentions’ of (7b) can be made quite precise in terms of set membership. For instance, one could identify the G of (7b) with an arbitrary Turing machine defined as in (8): (8)

Partee, ter Meulen, and Wall (1993: 508) “A Turing machine M is a quadruple (K, S, s, d), where K is a finite set of states, S is a finite set (the alphabet) containing #, s e K is the initial state, and d is a (partial) function from K × S to K × (S ∪ {L, R}).”

Then to say that G mentions some minimal form F is to say that F either is a member of K or is one of the symbols in the “situations” defined by the partial functions. More perspicaciously perhaps for linguists, one can take advantage of the equivalence of Turing machines and Type 0 grammars (Partee, ter Meulen, and Wall, 1993: 514) and take “mention” to mean that F is one of the nondesignated (e.g., not the arrow) symbols of one of the rules of a Type 0 grammar (that is, an unrestricted rewriting system). To illustrate the notions, consider that infinitesimally tiny but infinite subpart of English whose initial members are represented by (9): (9)

a. b. c. d. e.

My My My My My

father died. father’s father died. father’s father’s father died. father’s father’s father’s father died. father’s father’s father’s father’s father died.

Evidently, the collection of which (9) specifies an initial sequence consists of all and only those sentences of the form (10), where (X)* is the so-called Kleene star notation, which denotes any finite string composed exclusively of occurrences of the symbol string in parentheses in that order: (10)

My + (father+’s)*+ father + die+ ed

A partial grammar of this collection might be given by a totally ad hoc and unprincipled phrase structure grammar of the form (11): (11)

a. b. c. d. e. f.

S → DP + Verb DP → my +GenDP + father GenDP → GenDP + GenDP GenDP → DP + ’s DP+s → father + ’s Verb → die + ed

178


Now, assuming that the sentences in (9) consist of all and only the minimal forms my, father, ’s, die, and ed, it is clear that every minimal form in the collection (9) is mentioned in (11), in that my and father are among the symbols of rule (11b), ’s is one of the symbols of rule (11d), and die and ed are among the symbols of rule (11f). Let us say then: (12)

An NL is closed if and only if a. there is some finite grammar of NL = GNL such that b. every minimal form of every sentence of NL is mentioned in GNL.

A still simpler, essentially equivalent way of saying this would be (13a): (13)

a. An NL is closed if and only if there exists some finite list of all the minimal forms of every sentence of NL. b. An NL is open if and only if it is not closed.

Clearly, for any collection of sentences, hence any NL, to be enumerable, minimally it must be closed, for devices that recursively enumerate sets, that is, Turing machines or their equivalents, are self-contained. They can only compute an output member of the set to be enumerated by combining elements from a given fixed list, as in claim (6). For (3d) to be true then, it is necessary that NLs be closed.10 Evidently, the subpart of English specified in (9) is closed. But this tells us nothing about English as a whole. The essence of the question begging involved in claim (1a)/(6) necessary to justify (3d) is then that its author has never argued that full NLs are closed; nor has anyone else to my knowledge. In what follows I try to show that this logical gap in any argument for (1c)/(3d) cannot be filled for the simple reason that NLs are open, moreover, richly open, in several distinct ways.

4. Direct speech In the 1951 science fiction film The Day the Earth Stood Still, a recently arrived space alien (played by Michael Rennie) at one point speaks to his giant flying saucer defense robot (named Gort). One could report that linguistic event as in (14): (14)

The space alien said ‘klaatu barrada nikto’ to Gort.

Example (14) is one of endlessly many sentences of a perfectly well known variety. These contain a complement of a familiar type called direct discourse or direct speech. It is a standard and quite traditional notion with a traditional Latin name. So Jespersen (1924: 290) indicates: “When one wishes to report what someone else says or has said (thinks or has thought)—or what one has said or thought oneself on some previous occasion—two ways are open to one. Either one gives, or purports to give, the exact words of the speaker (or writer): direct speech (oratio recta).” Trask (1993: 83) defines direct speech as “[t]he reporting of what someone has said by quoting her/his exact words, as in ‘What time is it?’, she asked . . .” The difference between (14) and Trask’s


179

example is that the former purports to report the exact words of an instance of performance in a (presumed for argument) NL distinct from that of the report. In view of this difference, let us informally distinguish three types of direct speech: (15)

a. a domestic direct speech segment (DDSS): a piece of reported speech that is purported to be in the same NL as the containing report. b. a foreign direct speech segment (FDSS): a piece of reported speech that is purported to be in an NL distinct from the containing report c. a nonlinguistic direct speech segment (NLDSS): a piece of reported noise that is not purported to be in any NL.

Category (15c) is intended to allow for reports of, for example, noises by animals, noises of inanimate origin, tornadoes, explosions, trains, squeaking doors, and so forth. Consider now a purported constructive grammar GEnglish of one of the multitude of NLs referred to as “English,” say, my English, in which (14) might be a sentence. Since (14) is by assumption an English sentence, then if (3d) is true, GEnglish generates (14) and every other sentence of English, and if so, as already argued, English would have to be closed. The problem is that any derivation that involves any computational procedure/ Turing machine, for example, that postulated by Chomsky, will fail to yield (14) unless it is able to find the FDSS klaatu barrada nikto (or all of its components if any; see later discussion) in LXEnglish. Keeping doctrine (3d) consistent with data like (14) will apparently (but see later discussion) then require at a minimum maintaining the disjunction in (16): (16)

Either: a. (14) is not an English sentence, or b. The FDSS ‘klaatu barrada nikto’ (or its components [if any]) is listed in/an element of LXEnglish, that is, is mentioned by GEnglish.

Consider first (16a). This would follow from the more general: (17)

(14) is not an NL sentence.

The problem with any attempt to maintain (16a) is this: Surely if, contrary to (17), (14) is an NL sentence, it is an English sentence and not, for example, a Turkish or space alienese one. However, (14) seems quite well formed in my English. There is no intuitive fact to ground any rejection of its grammaticality. One does not teach children or foreigners to avoid such constructions. One does not correct such examples. Even the most extreme purist who bemoans linguistic deterioration would never cite examples like (14) as evidence. Popular grammarians never point to such examples as part of their prescriptive enterprise. And as already seen, traditional grammatical discussion recognizes their existence and assigns them to a type (oratio recta). There is then no direct factual basis for denial that (14) and similar examples are part of English. Thus (16a) is groundless.

180


One can then turn to (16b). If doctrine (3b) has any content, then a straightforward interpretation of (16b) is evidently not tenable. For all of (18a–f) and the endlessly many other cases relevantly like them have exactly the same status as (14): (18)

a. b. c. d. e.

The space alien said ‘slatu niraba miktu’ to Gort. The space alien said ‘tlato sniraba fiktu’ to Gort. The space alien said ‘drato zimboto shiktu’ to Gort. The space alien said ‘snato jikmoti puroboton yablotofoo korodor’ to Gort. The space alien said ‘grato shilt buzu ftmakvrss muktmwik rabsidobad vagomasitor’ to Gort. f. The space alien said ‘vngmssptkfookytz’ to Gort.

If the notion of fixed, finite lexicon is to have any substance at all, that is, if it is genuinely to be part of a computational procedure, the FDSSs of (14), (18), and so on, clearly could not be elements of an included lexicon. For an interpretation of (3b) under which it is consistent with the inclusion of all of these forms is an interpretation under which the notion ‘fixed lexicon’ excludes nothing. A ‘lexicon’ needed to cover (14), (18), and so on, could hardly even meet the minimal condition of being finite, given a lack of length bound on pieces of direct speech; see Langendoen and Postal (1984: chapter 3) for a general discussion of the nonexistence of length bounds of any sort in NLs. So nothing precludes distinct, arbitrarily long FDSSs or NLDSSs, of, for example, the form: (19) a. The foreigner screamed ‘neeeeeeeeeeeeeeeeeeeeeeeeeeeeee. . . .’ b. The cow went ‘moo’/‘mooo’/‘moooo’/‘mooooo’/‘moooooo . . .’

That is, there is no viable reason to believe that while, for example, English permits reporting as in (19a) human expressions or as in (19b) cow noises of various lengths, at some point there is a human or cow noise too long to be reported. What rule of English fixes a maximum length on reportable noises? Since it is impossible to specify an actual bound, the answer must be “none.” In short, and for several reasons, it is then impossible for every piece of direct speech, every DDSS, FDSS, and NLDSS, to be mentioned in the grammar of any NL. The points just made were in essence, as I am grateful to Geoffrey K. Pullum for reminding me, noted by Harris (1979) in remarks that subsequent linguistics seems to have ignored (and so, inter alia, never refuted); note the failure of Chomsky’s major linguistic works subsequent to Harris (1979), for example, Chomsky (1981, 1986a, 1986b, 1988, 1995b, 2000c), to even reference it. (20)

Harris (1979: 10–12) “Hence the set of sentences, as sequences of elements in a finite discrete set, is denumerably infinite, even though it will be seen below that the matter is complicated by the fact that the set of sentences is not well-defined and is not even a proper part of the set of word sequences.”

Harris inserts here a footnote, his 11:


181

The latter is due to the fact that there are sentences which contain sound sequences that are not words: Any sound can be the subject of a sentence of the form X is a sound, X is his name, X1 and X2 are different sounds even though we cannot hear the difference (5.4), etc. The set of objects that occupies the positions of X here, and so the set of sentences of the above forms, is not discretely differentiated (aside front the limits of discrimination of hearing and perception) and not necessarily denumerable.

Harris makes it clear here that (i) direct speech expressions can in principle be parts of grammatical sentences, that is, grammaticality will fail only when the non– direct speech portions are improperly formed; and (ii) that for that reason alone, the collection of sentences is not well defined and does not form a recursively enumerable set. He also raised the possibility that the NLs were thereby not denumerable sets but hedged with “not necessarily.” Implicit in Harris’s remarks is, of course, the claim that NLs are not closed. While Harris’s insightful claims are certainly correct as far as they go, they are too terse to really do justice to the major implications they have. He did not address various indirect moves to which a defender of the closed status of NLs might attempt to appeal. For despite what has been implied so far, there might seem to be a way of keeping the grammaticality of even infinitely many pieces of embedded direct speech consistent with a finite lexicon. This might appeal to some idea like the vague, undeveloped, repeated claim by Chomsky that an NL grammar assigns a structure to “every possible relevant physical event”: (21)

a. Chomsky (1986b: 26): “[T]he I-language assigns a status to every relevant physical event, say every sound wave.” b. Chomsky (2000c: 79): “[I]t could turn out that it assigns a specific interpretation to every possible signal.”

The application to the issue I am discussing here would be something like this: While it could be granted that there are infinitely many pieces of direct speech, and that these occur as parts of well-formed NL sentences, it could be claimed that (i) these pieces of direct speech are themselves recursively enumerable because (ii) every piece of direct speech merely involves some finite combination drawn from a fixed finite phonetic alphabet. An idea like this has already been proposed in passing by Green (1985: 123–124).11 These assumptions would permit taking the direct speech piece in, for example, (14) to consist neither of an arbitrary single unit nor of a sequence of, say, three such units. Rather, every direct speech sequence would consist of a fixed string of phonetic elements and every such string could be generated by a trivial infinite grammar finitely schematizable via the Kleene star notation as in (22): The notion ‘schematizable’ here can be given a precise interpretation in terms of metagrammars, computational devices that generate infinite collections interpreted as grammars. For this conception of finite grammars that generate nonfinite grammars, see Langendoen (1976). (22)

Direct Speech → (Phonetic Segment)*

182


This assumes the existence of some grammatical category Direct Speech and a recursively enumerable set of phonetic segments. The latter assumption might be claimed to have a certain plausibility if one limits its application to direct speech that involves purely linguistic performances. However, it is by no means obviously true, even for this class. The limited view itself is challenged in Pullum (1983). If there is such a thing, though, then the claim could be that the computational grammar would specify all and only the positions where the category Direct Speech could appear, for example, after verbs like say, whisper, yell, and so on, and after nouns like sound, utterance, form, expression, and such. It would then be necessary to expand the notion of lexicon to include not only the listed minimal forms but the entire computational output of (22) plus the principles that spell out the fixed set of phonetic segments. This would no doubt induce complications in particular versions of the doctrine under attack here. Viewing, for example, klaatu barrada nikto as made up syntactically of each of its phonetic segments seems entirely artificial. Is one, for instance, to maintain claims such as that all branching is binary by asserting that this has a huge number of alternative binary bracketings?12 Further, how are these putative syntactic constituents to be analyzed so as to be consistent with other putative universal constraints on tree structures, for example, X bar theories? In addition, normal syntactic constituents have meanings, but analyzing direct speech into syntactic constituents the size of phonetic segments precludes any assignment of meanings to these constituents. Finally, it is arguable that entire direct speech segments have meanings, as discussed in section 9, and if this is so, but their syntactic components do not, then claims about compositionality also run afoul of these cases. The obvious argument for the claim that direct speech segments have meanings is that, for example, (23a, b), (23c, d), and even (23e, f) (see [23g]) embody distinct propositions, a key point to which I will return: (23)

a. b. c. d. e. f. g.

Elmer grunted ‘you’ll never take me alive’. Elmer grunted ‘you’ll never make me a hive’. Ellen snarled ‘snedo’. Ellen snarled ‘fneto’. The cow went ‘moooo’. The cow went ‘moooooooooo’. Don’t be silly! The cow went ‘moooo,’ not ‘moooooooooo’.

But despite these problems, it might still be concluded that any notion that direct speech clashes fundamentally with the claim that NL grammars can be computational has been successfully gotten around. But even were this approach capable of rendering direct speech compatible with the idea that NL grammars can be proof-theoretic in the case of DDSSs and FDSSs, it still seems hopeless for NLDSSs, since the view that there is a universal computational alphabet capable of representing every nonlinguistic noise that can be reported with direct speech seems impossible to take seriously. I will not belabor this point here since it is taken up in a more general context in section 6 and, as considered in


183

the following sections, the idea of a purely computational account of performance reports faces distinct and, if anything, worse problems.

5. Gestural performance Even if one ignores the direct speech reporting of nonlinguistic noises and accepts the possibility of a recursive enumeration of phonetic segments for spoken NLs along the lines of (22), this would not suffice to solve the problem of direct speech for a view that NLs are closed. The reasons involve first nonspoken NLs, for example, the multitude of signed NLs of the deaf. Even the author of (1), who in the past had written as if NLs and phonetically performed NLs were coextensive, as in (21a), has recently said: (24)

Chomsky (2000c: 121) a. “Though highly specialized, the language faculty is not tied to specific sensory modalities.” b. “Thus the sign language of the deaf is structurally much like spoken language.”13

Given, though, the fact that some NLs systematically use nonphonetic gestures to create tokens of their sentences, to maintain a claim that NLs as such are closed even in the face of direct speech, it cannot suffice for there to be a recursive enumeration of phonetic segments relevant only for the subgroup of NLs that are in standard circumstances performed via sound. Rather, under the assumption that there are analogs of direct speech constituents in nonspoken NLs,14 there would have to be as well a recursive enumeration of a gestural equivalent of discrete phonetic segments for every physical gesture that could underlie the performance of any nonspoken NL. Far from being plausible, this seems chimerical. That it is might well be implicit in the claim cited in Perlmutter (1986: 523) that pertains to signing: (25)

Whitney (1875/1979) “Among their manifold capacities, they are able to make gestures, of infinite variety, all of which are reported by the vibrations of the luminous ether to a certain apprehending organ, the eye, both of the maker and of others.” (emphasis mine: PMP)

In any event, anyone who claims that there is a recursive enumeration of the gestures capable of serving as parts of performances of all signed NLs bears the heavy (and to my knowledge never assumed) burden of supporting such an idea. Second, it can be argued that the demarcation between phonetically expressed NLs and others like the gesturally signed NLs of the deaf is incomplete. This means that the logic that shows how gesturally expressed NLs attack the notion that NLs are closed can be applied internal to standard phonetically expressed NLs. That is, I claim that even basically phonetically expressed NLs like English allow restricted gestural and other nonvocal forms of expression, specifically in certain direct speech and related contexts. So there is every reason to take, for example, (26a and b) to be

184


schemas for endlessly many English sentences whose next to final constituent is gesturally signed: (26)

a. The deaf person went ___ yesterday. b. The deaf person made the gesture ___ in the living room.

To perform a relevant instance of (26a) one articulates the first four words and then makes at that point (corresponding to the dashes) a gesture appropriately identical to that which is being reported, then pronounces yesterday. Other examples of the same type that have nothing to do with NLs associated with the deaf include (27b), a truth functional equivalent of (27a): (27)

a. When the cop told her to leave, Sheila gave him the finger (twice). b. When the cop told her to leave, Sheila went ___ (speaker makes the appropriate gesture) (twice).

So, barring a factually unmotivated claim that instantiations of schemas (26a, b), (27b), and so on are not NL sentences (e.g., sentences of English in this case), to maintain Chomsky’s claim (3b) requires again the posit of a recursive enumeration of gestures.

6. Beyond gestures I have so far considered the implications for claim (3b) of types of NL sentences that involve direct speech and nonphonetic gestural performance. But the situation for a defender of NL closure is more threatening than that entailed by these phenomena. First, as already touched on, there are NL sentences that bear certain similarities to direct speech structures but involve reference not to linguistic performance, that is, in the standard case, to utterances taken to perform sentences, but to other sorts of noises, either vocally produced or not. For instance, (28a) might be a schema of descriptions of the noise made by a person afflicted with serious snoring, while (28b) might schematize the description of the noise associated with a tornado: (28)

a. He goes ___. b. It gives off a roar like ___.

The relevance of such cases is that even though the material schematized by the blanks in (28) involves proper performance via the creation of sound waves, there is, evidently, no reason at all to imagine that the full range of such performances is coded by anything like a universal phonetic alphabet. That is, there is no reason to believe that, for example, the class of examples illustrated in (29) that purport to indicate bump-induced car noises is a priori specifiable in a linguistic way: (29)

Pullum and Scholz (2001: 17) My car goes ‘ehhrgh’ when I go over a bump.


185

This position is consistent with the observation of Kathol and Levine (1993: 210, note 7): “Thus inarticulate cries, imitations of animal or industrial noises, indeed anything producible by the human vocal tract can appear within the fronted quotation.” The point is strengthened if, as, I would claim, it is correct to see such schemas as (28) and the quoted part of cases like (29) as covering NL sentences where the blanks or quoted material are performed even without human vocal apparatus, for example, by clapping, or utilizing an arbitrary mechanical means of producing sounds, orchestras, guitars, machine guns, or whatnot. That is, I suggest that while it is certainly proper to perform (29) by making a vocal noise after the word goes, it is just as proper to perform it by playing a recording of actual car noises in that position and just before one pronounces when I go over a bump. If so, any possibility of reducing such cases to consistency with (3b) via analogs of (22) is out of the question. Support for the view that NL sentences can involve nonvocal apparatus noise is provided by the remarks in (30) about metalinguistic negation: (30)

Horn (1985: 136, note 12) “As Barbara Abbott has pointed out to me, u need not even be a specifically linguistic utterance, as seen by the function of metalinguistic negation in the following musical scenario: Piano student plays passage in manner µ. Teacher: ‘It’s not [plays passage in manner µ] __ It’s [plays same passage in manner µ'].’”

Here Horn’s µ, µ' denote some physical characteristics of piano-playing performances. While Horn does not assert that the teacher has performed NL sentences, using only the term scenario, his discussion only makes sense on that assumption. For he is taking the examples to support his view about the nature of metalinguistic negation. This would be illogical unless the performances in question were instances of performance of NL sentences. If the entities involved were not NL sentences, that is, not part of the NL being discussed, how could they positively support a claim about the way metalinguistic negation is to be analyzed in that NL? That is, if no NL sentences are involved in the scenarios, they could no more support a claim about metalinguistic negation than, for example, (31a, b) can support a claim that English has verb final transitive clauses: (31)

a. *Melissa will her roommate cheat. b. *The government should the elderly support.

Second, I see, moreover, nothing in the structure of NL itself that limits cases like (28) to sounds. For instance, the written medium can be seen to bring out the fact that, again in a regimented set of contexts, even regularly phonetically signed NLs allow graphic or geometric realizations of certain constituents. Thus I claim that (32a and b) schematize endlessly many English sentences, where the blank denotes some shape; one of these can be represented on paper as in (32c):

186 (32)


a. The professor drew ___ on the blackboard. b. The professor drew a figure of the form ___ on the blackboard. c. The professor drew a figure of the form P on the blackboard.

Each distinct instance of (32a) can be performed by articulating the first three words, then providing some representation of the geometric form and then pronouncing the final three words. For (32b), one pronounces the first eight words, then somehow instantiates the figure and then pronounces the final three words. Hence one instance of (32a) will be truth functionally equivalent to (33): (33)

The professor drew a circle with the letter P in it on the blackboard.

Actual instances of such sentences are found, even in the linguistic literature. Here is one from a recent monograph: (34)

Culicover (1999: 28) “Conversely, the fact that hotdog means

is not predictable from ‘hot’, ‘dog’ or the combination.”

I see no reason beyond a priori dogma for denying that Culicover has as much instantiated an English sentence with (34) as he did with his following remark, which incorporated no image. To avoid an arbitrary claim that (32a and b) do not represent actual NL sentences and that (32c) and (34) are not written representations of actual sentences, maintenance of (3b) would then require not only a recursive enumeration of gestures but also an enumeration of all the geometric forms that could be covered by (32a and b). Given the uncountable character of even the collection of all planar (two-dimensional) forms, whose number is of the order of the real numbers, this is impossible.15 Moreover, any restriction to just two dimensions seems artificial, since (35a) could be a truth functional equivalent of one instance of (35b): (35)

a. The sculptor carved something of the form of a cube out of sandstone. b. The sculptor carved something of the form ___ out of sandstone. c. The device produced a three-dimensional image just like ___.

To perform a relevant instance of (35b), one pronounces the first seven words, then displays a cube, then pronounces the last three words. The way to perform an instance of (35c) is to pronounce the first eight words and then to in some way introduce the relevant image, for example, by turning on multiple slide projectors, by displaying a holograph, by holding mirrors in a certain way, and so on. But, since


187

the number of three-dimensional objects is, via the logic of note 15, of the order of real numbers, it follows that the class of constituents schematized in (35b, c) is also of that order.

7. Metalinguistic structures A type of linguistic information-representing sentence distinct from all the various ‘direct speech’ varieties is relevant to questions of openness, a type represented by, for example, (36): (36)

a. The French word for milk is ‘lait’. b. To express ignorance of some topic, one can say in French ‘va savoir’. c. German ‘Kopf’ is equivalent to French ‘tête’.

Such examples, like standard direct speech representation cases, contain parts that seem to involve NL elements. The difference is that a direct speech sentence purports to describe a particular performance; the direct speech report references an object with space/time coordinates. The foreign parts of examples like (36), however, are not putative descriptions of any performance of the NLs referenced. They purport instead to denote elements of those NLs themselves. Let us refer to the relevant parts of such examples as metalinguistic constituents. Clearly, examples like (36) that contain metalinguistic constituents are common and intuitively entirely grammatical. Any denial of their well-formedness would be an act of desperation. Notably, those whose theoretical position might demand such a rejection utilize metalinguistic examples like (36) en masse, without the slightest indication that they are in any way abnormal. Some examples from Chomsky (1988), which contains by a rough count more than two hundred and forty others, are displayed in (37): (37)

a. “The verb examinar requires an object, . . .” (page 95) b. “One of the traces must be bound by nos and the other by al que.” (page 96)16 c. “Thus (2) must be understood in the manner of (8a), not (8b) (where lo stands for el hombre).” (page 98) d. “Let us first take the case in which the clitic attaches to afeitar, forming afeitarse.” (page 86)

Denial that (37a–d) are English sentences would have such implications as that touched on in section 1, namely, that it would not be possible to express the linguistics of NL1 in any NL distinct from NL1. Given that metalinguistic expressions occur in grammatical sentences, the question arises of how they can be kept consistent with claim (3b). The same pattern of argument involved in the discussion of direct speech segments becomes relevant. While the infinite number of metalinguistic forms obviously cannot be listed in any LXi, it might be claimed that metalinguistic forms nonetheless fall within the domain of recursive enumerability via appeal to an analog of (22) something like (38):

188


(38) Metalinguistic Constituent → (Phonemic Segment)*

The choice of phonemic rather than phonetic segments here seems natural (see [2]) but is not crucial. The same artificiality issues would arise with respect to taking the syntactic structures of metalinguistic stretches to be phoneme-sized. More important, the same issues of gestural forms, geometric forms, and such attack any claim that a universal theory of phonemic segments could cover all metalinguistic constituents. For instance, this is hardly conceivable for instances of schemas like (39): (39)

a. In the NL of the deaf of Gwambamamba, the gesture ___ means “why not?” b. In America, the gesture ___ means “screw you.”

Overall then, metalinguistic sentences just strengthen the arguments against (3b) from direct speech structures, gestural structures, and so forth.

8. The controversial constituents as real constituents One can suspect that there will exist considerable resistance to accepting as real NL sentences the sort of wholly or partially nonphonetically performed objects I have claimed are NL sentences, which include those schematized in (26), (27b), (28), (29), (30), (32a, b, c), (34), (35b, c), and (39). Part of such resistance might just be the traditional association of sentence with its pronunciation, reinforced by frequent repetition of remarks like those in (40): (40)

a. Chomsky and Halle (1968: 3) “The grammar of the language is the system of rules that specifies this soundmeaning correspondence.” b. Chomsky (1972c: 11) “A person who has learned a language has acquired a system of rules that relate sound and meaning in a certain specific way.”

Besides what can now be seen to be an exaggeration of the link between NL sentences and sound performance in particular, such statements are misleading in another respect. Actual sound can only be produced by a physical object capable of producing a physical disturbance in a medium like air, in this case the speech apparatus. An NL itself cannot directly link structure and sound, since actual noise can only be produced by performance, by setting some physical objects in motion. What an NL can do, and for the most part does,17 as made clearer in Chomsky (1975a: 18), is link various abstract structures, which include a superficial one that can be interpreted (according to some sort of conventions) as instructions to a certain physical apparatus. From this point of view, the interpretation conventions that associate particular parts of sentences with physical instantiations of particular kinds might be regarded as external to the NL proper. In any event, the interpretation via articula-


189

tion, nonvocal gestures, or whatever, is not determined by the abstract structure. Given that, there is no reason to assume an inherent relation between NL sentences and sound in particular. On the positive side, one reason to view the sort of constituents at issue here as real NL objects is because they seem to have key properties of uncontroversial constituents.18 For instance, all the examples cited so far had the form of independent (in fact, declarative clauses) or simple embeddings. But this is not at all necessary. Such structures can yield questions, commands, and suggestions, can appear as restrictive relative clauses, as complement clauses, as parts of predicate clauses, and so forth, as illustrated in (41)–(43): (41)

a. b. c. d. e. f.

Did the alien shout ‘klaatu barrada nikto’? Don’t go around whispering ‘klaatu barrada nikto’. I suggest that you never grunt ‘klaatu barrada nikto’ at that robot. Every alien who shouted ‘klaatu barrada nikto’ was executed. They reported that the alien shouted ‘smato marada snikto’. The right thing to do was to scream ‘klaatu barrada nikto’.

(42)

a. b. c. d.

Every deaf person who went ___ was arrested. When did Marsha make the gesture ___ with her right hand? The facial expression ___ often indicates anger. They reported that she made the facial expression ___.

(43)

a. b. c. d.

No professor who drew ___ on the blackboard was rehired. The sculptor who carved something of the form ___ was criticized. A star shaped like ___ is associated with the Jewish religion. They denied that she drew ___ on the blackboard.

Moreover, constituents of the relevant kind can be coordinated and negated: (44)

a. b. c. d. e. f. g. h.

The alien shouted ‘klaatu barrada nikto’ and not ‘slatu niraba miktu’. What did the alien shout? Not ‘slatu niraba miktu’. Carla went ___ or ___. The alien shouted neither ‘slatu niraba miktu’ nor ‘smatu birada smakto’. The ASL/English bilingual went ‘please have some pie’ or ___ twice. The sculptor carved things of both the forms ___ and ___. He drew not the shape ___ but rather the shape ___. The teacher played not __ on the piano but ___ on the tuba.

Further, constituents with gestural or other nonsound performances can be antecedents for ellipsis: (45)

a. The space alien went ___ with his right tentacle, but I didn’t ___ with mine. b. Male space aliens can go ___ with their tentacles faster than female space aliens can ___ with theirs.

190


c. Although Mercedes drew the shape ___, I didn’t. d. While Noriko can’t carve a three-dimensional image of the form ___ on Friday, she can on Saturday.

In such cases, it is clear that the ellipsis-containing constituent involves a claim about the same nonverbally performed constituent as the antecedent.19 And such nonverbally performed constituents can also be the antecedents for non-null anaphoric elements, highlighted in (46): (46)

a. The space alien produced an image just like yours, but I did not produce an image anything like that. b. The space alien produced an image just like ___, but I did not produce an image anything like that. c. Sheila gave the cop the finger, which is a vulgar gesture. d. Sheila went ___, which is a vulgar gesture. e. Sheila gave the cop the finger but I did not make that gesture. f. Sheila went ___, but I did not make that gesture. g. Sheila went both___ and ___, but Glen did not make the latter gesture. h. *I did not make that gesture, but Sheila gave the cop the finger. i. *I did not make that gesture, but Sheila went ___.

Given that performances of such cases yield no sense of grammaticality, seem to have the standard properties of uncontroversial sentence performances (note the parallel ungrammaticality of [46h, i] for violating a constraint that evidently does not allow the anaphoric device to both be in a coconjunct of the conjunct that contains the antecedent and to precede that coconjunct), and provide no general obstacle to formulating a coherent account (see section 10), it seems that their inclusion in NL is deniable only on purely doctrinal grounds.

9. The irrelevance of the historical dynamics of lexicons A certain range of well-known and undisputed NL phenomena have not been cited here as objections to (3b) and should not be. I refer to the omnipresent existence of lexical accretion and loss, which over time add new forms to, and more slowly remove old ones from, the lexicons of NLs continuously. So fifty years ago there was no English form laptop referring to a type of computer, no verb suck meaning “to be of minimal value/quality,” and so on. Just so, previous to this work the forms open, openness, and so on were not used in the way they are here. But according to Partridge (1970: 153), before 1915 there was a word chuck-up meaning a military salute, a form I had never heard of and which evidently is not part of current English. When one looks at the full range of such phenomena, it might seem that one is also dealing with facts that render the notion of a fixed lexicon incompatible with linguistic reality. However, the dynamic sociohistorical processes that alter historical lexicons in this way have nothing to do with the point being argued for in this study, that is, have


191

nothing to do with the issue of whether NLs are open or closed. The reason is that such historical phenomena can be viewed as mappings from one finitely specifiable lexicon LX1 to another LX2 such that if each is combined with the remaining elements of an NL, the result consists of two (relatively trivially) distinct NLs.20 This is entirely consistent with the view that NL variants that differ in lexical elements, like those in (47), also represent trivially different NLs: (47)

a. British: The bonnet of the Jaguar was scratched/wants washing. b. American: The hood of the Jaguar was scratched/needs washing.

Such variation also is irrelevant to the issue of whether NLs are closed. But lexical change is in no way parallel to the phenomena illustrated by (14). While the former can be viewed as a historical phenomenon that really amounts to the instantiation of (relatively) trivially distinct NLs, it is absurd to imagine that there is a pre-1950s English that excludes the expression klaatu barrada nikto and a post1950s one that contains it. The absurdity of the assumption is fully revealed by, for example, the infinite open-endedness of the example collection illustrated in (18). That is, direct speech and the related phenomena discussed previously do not involve historical processes of NL change but rather for every NL at every point the possibility of representing infinitely many unconstrained physical performances. Just so, a “dynamic” approach equally lacks any application to cases like (37), since it makes no sense to imagine that every sentence that involves a descriptive remark about a hitherto unmentioned foreign form in NL5 represents a historical modification of NL5.

10. Unregimented constituents and the nature of NL sentences Standard views of NL sentences, fairly represented, I believe, by (1a) and (6), are arguably correct over a certain range of sentences. But such views fail, as argued in earlier sections, when faced with the variety of NL direct speech sentences, metalinguistic sentences, gesturally represented sentences, sentences that incorporate pictorial, geometrical or multidimensional objects, and so on. The standard view has thereby deeply underestimated the richness and expressive potential of NLs. In this section, I wish to show how this richness of expressive power can be given a coherent interpretation internal to what I believe is independently the only viable ontological view of NL sentences, namely, that they are abstract objects and, more specifically, that each NL sentence is a set.21 I want to suggest that the line between sentences that can be viewed as based on a fixed, finite lexicon, as in (1a), and those that cannot corresponds to several types of division of constituents. The first distinction is that between constituents of NLx that are, as (1a) requires, wholly based on forms mentioned in the grammar of NLx. I will call these lexically pure constituents. So, in (48), the subject constituent is lexically pure, but the object constituent, the verbal phrase constituent that contains it, the whole clause, and so forth, are not. Call them alexical constituents:

192 (48)


The newly arrived alien shouted ‘vlaatu worrada smeikto’ at the mailbox.

A second distinction involves the relation between the abstract objects, in fact, sets, that are constituents and any physical events, objects, and such that serve as genuine tokens of constituents. The basic idea is this: One class of constituents, those normally considered in linguistics, contains specific elements, in the standard case a phonetic representation, which can be regarded as a recursively enumerated object capable of being interpreted as instructions to a fixed physical apparatus. I will refer to such constituents as regimented. There is little need to say more about them here. But part of the force of earlier sections can be summed up simply by saying that NLs incorporate certain constituents that are unregimented; these include direct speech constituents whose performances seem to include arbitrary noises, image constituents whose performances seem to include arbitrary two-dimensional representations, and so on. And the current task is to see how one can make sense of unregimented constituents, specifically in context with the view that sentences are abstract objects of the type set. Arguably, of the four types of constituents logically constructable from the two distinctions just made, only three are characteristic of NLs. The normally considered constituents are both lexically pure and regimented. While all of the theoretically unusual constituents that have been considered in earlier sections are alexical, many are naturally taken as regimented. Others though, like those schematized in (32b), (35b), and so forth, are clearly not. What I see no way to instantiate, though, is a constituent that is both lexically pure and unregimented, and I assume that there are no such objects. I therefore concentrate on the question of how to give a theoretical account of the notion of a constituent that is both alexical and unregimented. It has been claimed that sentences are sets, and it is natural and I think correct to assume that sentences have a complex set-theoretical structure, which involves sets with other sets as members as well as some sets whose members are whatever the appropriate primitives for characterizing lexically pure and regimented constituents turn out to be. The question then is how alexical, unregimented constituents can fit into such a set-theoretical framework. The only answer I see is that an alexical, unregimented constituent C must represent a set whose elements include the physical tokens that make up individual performances of C. Let us clarify this through maximally simple examples: (49)

a. Felicia yelled ‘smekto’. b. Felicia yelled ‘smektof’.

The crucial property such examples have from the point of view of maintenance of the conception of sentences as sets is that distinct elements seem to be defined by distinct performances, more precisely by sets of performances. So (49a and b) represent distinct propositions and hence, if one assumes that meanings are parts of sentences, distinct sentences. This entails that the sets that comprise NL sentences must be able to contain as members or submembers something that can instantiate the endlessly distinct physical properties involved in direct speech. The only way I see that this can be the case is if direct speech segments involve sets that contain the


193

physical properties themselves and not, as in the case of more standard (regimented) linguistic elements, symbols that represent instructions (to a fixed physical apparatus) to produce physical things. What I am claiming is that the object constituent of the unique English sentence represented by (49a) is a set one of whose subsets contains as elements the actual physical sound waves produced by performances of (49). Note that a priori there is nothing strange in talking about sets, abstract objects, at least some of whose elements are physical objects, that is, nonabstract objects. This is common in introductory discussions of set theory. So Halmos (1960: 1) says: “A pack of wolves, a bunch of grapes, or a flock of pigeons are all examples of sets of things.” And Allwood, Andersson, and Dahl (1997: 3) indicate: “We might for instance choose to consider the set which consists of the Premier of Sweden, the smallest moon of Mars and the square root of 7.” So if there are three marbles on a table, one can speak of the set that consists of exactly those three marbles, and I can speak of the set of all electrons in the universe and ask questions about it, for example, about its cardinality.22 The fact that the latter might be currently unspecifiable in no way attacks the existence of the set in question. Just so in the case of the unregimented object constituent in (49a) the fact that there is no way to specify the size of the set of sound waves that make up all its tokens in no way argues against the existence of such a set.23 Taking an unregimented constituent to involve a set of its tokens is evidently superior to taking it to involve some single, specific token. The latter is incompatible with the fact that no less than regimented constituents, unregimented ones have unlimitedly many potential tokens. So, for example, (49b) can be said repeatedly and no particular sound wave associated as a token of its object has any primacy with respect to the specification of the nature of the object constituent. The idea that an unregimented constituent involves a set of its tokens receives some support, I believe, when one considers the semantics of cases such as (49) that involve them. I claimed earlier that direct speech constituents had meanings. It would perhaps be better to say “have denotations.” For, under the account of unregimented constituents just sketched, these denotations are, in general, informal terms, not hard to specify. The idea is as follows: An ordinary referential constituent, for example, the one associated with that dog, has a denotation that is independent of any token of that constituent. But not so for unregimented constituents. For these, the denotation is one of the set of associated tokens: (50)

The denotation of an unregimented constituent X one of whose subsets is a set of physical tokens T is some member of T.24

So if one specifies that the object constituent of (49a) is defined by a set of physical tokens T1, an utterance of (49a) by someone at 10 P.M. on October 31, 2002, amounts to the instantiation of a true proposition only if at some earlier point in time Felicia produced a yell that was a sound sequence S, where S is a member of T1. While what I have just sketched seems to me to be correct, it does have, I grant, one aspect that might seem like sleight of hand. It is all very well set-theoretically to speak about unregimented constituents being subsets of physical tokens. But since in general these tokens can be neither listed nor recursively enumerated, how are these

194


sets specified? One way to understand this question is to ask what grounds a claim that, for example, (51a and b) are distinct direct speech sentences, accounting, for example, for the sensible character of (51c): (51)

a. Helen whimpered ‘snedo’. b. Helen whimpered ‘sneto’. c. Helen whimpered ‘snedo’, not ‘sneto’.

It appears inevitable that unregimented constituents must be assumed to involve some kind of equivalence conditions, not necessarily (and surely not) the same for different types of unregimented constituents. Hence when one produces a token of an unregimented constituent C, one will specify the set that defines that constituent by virtue of the equivalence condition somehow associated with C. Arguably, these associations are derivable from the linguistic contexts in which unregimented constituents occur. For example, the fact that a constituent C is the object of the specific verb whimper, as in (51), might determine the relevant equivalence conditions. Since whimper describes a form of verbal performance, it is natural that these conditions would then involve some kind of identity of sound features but exclude as irrelevant such properties of verbal noises as those that define individual voices, which involve loudness and so forth, since these are in general properties irrelevant to claims about whether so and so whimpered such and such. Just so, the fact that an unregimented constituent is object of a verb that involves the creation of two-dimensional figures like that in (52) would naturally impose certain equivalence requirements and exclude others: (52) The professor drew ——— on the blackboard.

Here the equivalences would exclude sound but include properties of two-dimensional objects, geometric characteristics, color, and such. So the overall idea is roughly that the set of physical tokens taken here to characterize an unregimented alexical constituent C is specified as follows: Any fixed physical performance P interacts with the equivalence conditions EQ imposed by the context of C to define the token set that characterizes C. The logic is simply that the relevant set is the set of all elements X such that EQ holds between X and P. I cannot attempt here to say more about the required equivalence conditions. But nothing seems to emerge that interferes with the account suggested, in which an unregimented constituent involves a subset of its tokens, where the membership requirement involves crucially equivalence conditions imposed by the meanings present in linguistic context of the unregimented constituent. To conclude the discussion of the reality of NL constituents that are alexical and unregimented, it is worth mentioning a proposal of Partee (1973: 416). She suggested very briefly that a possible way of describing direct quotes and sentences that seem to contain gestures would involve recognition of invisible demonstratives plus a claim that the direct quote or gesture was actually not part of the sentence but only part of a larger discourse. The idea would then be that, for example, (53a) would be an elliptical form of (53b):


(53)

195

a. Morris went ——— (where the blank represents a vocal noise X). b. Morris went like this: ——— (where the blank represents a vocal noise X).

The connection between the demonstrative and the putative nonsentence part would then be mediated by some sort of contextual algorithm that picks out certain objects in or properties of the whole context as a referent of the demonstrative. Partee’s model for this type of description was, specifically, cases like (53a), in which the quoted material referred to by the demonstrative follows everything else in the sentence. This makes the sentence-external view of such at least conceivable. But of course, that property is not at all general and endlessly many cases are not subject to such a treatment: (54)

a. Everyone who claimed they went ——— may have really gone ——— or at least thought they did. b. Since the claim that she went ——— has not been refuted, I can assume that someone who went ——— was seen.

Moreover, on other grounds, Partee’s sort of account would have no obvious application to cases of the sort dealt with by Jackendoff (1984), discussed later, where no analog with a demonstrative is grammatical: (55)

a. b. c. d.

The sound ——— is grating. *The sound that is grating: ———. He discussed the gesture ——— on Friday. *He discussed the gesture that on Friday: ———.

11. Conclusions The principal result of the preceding sections has been an array of arguments that support the untenable character of the claims in (1), repeated here: (1)

a. Chomsky (1959: 137) “A language is a collection of sentences of finite length all constructed from a finite alphabet (or, where our concern is limited to syntax, a finite vocabulary) of symbols.” b. Chomsky (1959: 137) “Since any language L in which we are likely to be interested is an infinite set, we can investigate the structure of L only through the study of the finite devices (grammars) which are capable of generating its sentences.” c. Chomsky (1959: 138) “The weakest condition that can significantly be placed on grammars is that F be included in the class of general, unrestricted Turing machines.”

The key element of (1a) for our concerns here is the view that each NL sentence is constructed from a finite (hence listable) alphabet of symbols or a finite vocabulary of syntactically minimal forms, a fixed lexicon. Accepting (1a) amounts to ac-

196


cepting that NLs have the property I have called closed. It has been argued to the contrary that (1a) must be seen as false and NLs regarded as open because of diverse facts that involve direct speech, nonphonetic gestures, nonlinguistic noise, geometrical forms, metalinguistic constituents, and so on. All these phenomena reveal that NLs incorporate certain constituents whose minimal elements are not mentioned in any grammars; these are the alexical constituents of section 10. Several aspects of doctrine (1a) are worth highlighting. First, the logical conjunction of (1a, b, and c) defines the technical aspect of the generative grammar view of NLs. The untenability of (1a) thus undermines that general view. Second, (1a) was introduced in the 1950s with no supporting argument Third, it has been maintained by its originator and a multitude of those he has influenced ever since also with no supporting argument. Works like Chomsky (1995b) that advance the minimalist program of the 1990s and later and incorporate (1a) neglect to argue for it, just as did those of the 1950s. That the doctrine flourishes in the face of this lack suggests that many may have assumed that (1a) is a sort of self-evident truth. But it has been argued here that not only is (1a) not self-evident, but also unchallenged factual properties of NLs incompatible with it (e.g., the existence of various forms of direct speech, metalinguistic constituents) were known long before (1a) was enunciated. This principle seems in short to be nothing more than a (strangely) popular dogma, which is evidently an extraordinary state of affairs. If one were to have asked a priori whether a doctrine incompatible with well-known and traditionally discussed (e.g., by Jespersen, 1924) features of the domain of study could be introduced in the mid–twentieth century in a growing field and successfully take root and maintain itself over a period now approaching a half-century, one would have tended to answer in the negative. But the facts are otherwise. No doubt one reason that (1a) was so easily maintained in spite of its grave incompatibility with common fact is that the issue it raises was rarely discussed. A notable exception is provided by Hockett (1966: 182–183). Hockett ended up assuming (1a), specifying (1966: 183) as in (56): (56)

“Any sentence in the language is a string of characters, each of which is one or another of the characters of the fixed finite alphabet.”

However, this claim by Hockett was not question begging, since prior to that he had considered the factual adequacy of the claim, characterized it as open, and only specified (56) as a working assumption. Hockett’s discussion was phrased in terms of a striking and insightful implication of (1a) for the logical connection between NL and typewriters. He noted (1966: 182): (57)

“A point of departure for the formalization of language description is afforded by the fact that one can design a typewriter for any human language. . . . The keyboard and type bars give us a small, clearly defined stock of symbols among which we must make our choice for any key-striking. The carriage motion is such that our choices appear in a linear sequence on the paper. The operator can override these arrangements—I have


197

seen pictures drawn with a typewriter—but let us set aside such aberrations. If the stock of symbols is the correct one for a given language, then we can write practically anything we wish in that language. “Any linguist who uses a typewriter has had the experience of lacking some symbol he needs and having to improvise or to arrange for his keyboard to be modified. There are two possible interpretations of this. One is that a typewriter’s limitations are only a matter of economics. Symbols but rarely needed can more cheaply be written in by hand as the occasion arises, but in principle there is no reason why we should not incorporate everything we will ever need for a given language on a single keyboard. Of course, not everything we put on paper is language. If the supply of useful type faces is indefinitely large only because of charts, graphs, pictures, and other nonlanguage items, the first interpretation can stand. The second interpretation is that the limitation is not merely practical but essential. No matter how large we were to make the keyboard for a particular language, according to this interpretation, we might still encounter a need for symbols not provided on it. “It is not at all obvious which of these interpretations is empirically correct. It is clear, however, that for purposes of analysis and description the first interpretation is customarily assumed. If the assumption is in fact true, all is well. . . . Our assumption is, then, that if one has the right keyboard one can type, not just almost anything one wishes in a given language, but any sentence of that language. The keyboard supplies a finite alphabet of characters. Any sentence in the language is a string of characters, each of which is one or another of the characters of the fixed finite alphabet. On the other hand, not every string of characters drawn from the alphabet is a sentence of the language. The string of symbols that begins with the last capital letter before this and ends with the next period is a sentence of written English. The string that follows the next colon and runs to the following period is drawn from the proper alphabet, but is not, I believe, a sentence of written English: fkwwy qpat emff agvktom. Every sentence is a string of characters from the appropriate alphabet, but not vice versa.”

Hockett thus saw with full clarity that there was a factual claim involved in (1a). And his typewriter account permits one to see in a most graphic way the nature of the issue and why the evidence of earlier sections that involved, for example, arbitrary noises, arbitrary gestures, arbitrary pictures, and so forth, that is, unregimented constituents, shows that (1a) cannot be true. Moreover, Hockett’s description of the regimentation issue in terms of typewriters permits a novel and very vivid alternative way of specifying the nature of openness, as follows: (58)

a. T is an abstract typewriter/printer if and only if T consists of finite set of elements called keys, {k1 . . . kn}, each associated with some single symbol S(ki). b. A word string W is typeable/printable (T) (‘printable by T’) if and only if W consists of a finite string of symbols [Z] and every member of [Z] (except perhaps ‘space’) is a member of {S(k1), . . . , S(kn)}. c. An NL, NLx, is closed if and only if there is some abstract typewriter/printer T such that the word string of every sentence of NL is typeable/printable (T). d. An NL that is not closed is open.

198


What Hockett did not see then is that evidence that shows the correctness of his ‘second interpretation’ of the relation between NLs and abstract typewriters was easily at hand. No abstract typewriter provides the means to type/print every sentence of an NL that allows direct speech, gestural or pictorial constituents, and such. Any such NL will have indefinitely many sentences at least one part of which is unprintable. Moreover, in effect, a perception of the falsehood of (1a) was also reached by Jackendoff (1984). He was concerned with an English construction that I have not so far mentioned, one that he characterized (1984: 25) as consisting “of a definite article and a noun followed without pause by an expression E which can be of quite varied character; I will refer to it as the the N—E construction. Here is a range of examples, grouped approximately into semantic categories.” Jackendoff’s large list of examples included those of (59): (59)

a. the phrase the phrase b. the word/verb run c. the prefix un d. the construction N of NP e. the sentence Will you marry me f. the sequence up a g. the sound ph h. the syllable pa i. the letter A j. the number 14 k. the note E l. the noise ***** [raspberry, imitation of’ a goat, etc.] m. the pattern da-dum da-dum da-dum n. the symbol $

Crucial from my perspective here, of course, are cases like (59i, k, l, m, and n), where in our terms the expression E is alexical and in some cases (clearly, l, and n) unregimented. About such cases Jackendoff (1984: 26) concluded: (60)

“On the other hand, there are no inherent syntactic constraints on E: it need not be a syntactic constituent—as in (1); nor even an expression of English—as in (3). In fact, if the construction is uttered, E need not be expressible in standard orthography (as I have tried to suggest in [8a]); while if the construction is written, E need not have a pronunciation, as in (9). Hence, like the complements of verbs such as say and go (in the sense ‘make a noise’), E is a constituent whose interior is unconstrained by normal rules of syntax and phonology.”

Clearly then, Jackendoff had recognized that the constituents he called ‘E’ were overall unregimented and alexical and thus implicitly that no NL that contains an E-like constituent could satisfy (1a). He thus had in hand nineteen years ago a tool for showing further the entailed falsity of (1b, c). But, like Hockett eighteen years earlier, Jackendoff did not take this step.


199

Rather, he concluded for some reason that alexical, unregimented constituents could be subsumed internal to the generative apparatus of the time, arguing further: (61)

Jackendoff (1984: 26) “We will assume, therefore, that the phrase structure rule responsible for introducing E violates the normal theory of syntactic categories by permitting a totally free expression.”

But assumption (61) is impossible. For there can be no phrase structure rule or indeed any generative rule that permits a “totally free expression.” Because “totally free” is incompatible with the fact that E subsumes objects whose minimal elements are drawn from no list. But any generative rule ultimately specifies a class that is closed. Jackendoff’s remark might fairly be construed as a confusion that wrongly attributed to a putative generative rule properties that could only hold of a nongenerative, model-theoretic one (see later discussion). Suppose, for example, one offers (62) as a putative phrase structure rule that reconstructs Jackendoff’s idea: (62)

E → anything at all

The intention would be to have a rule that permits a “totally free expression.” But contrary to the intention, the characterization of phrase structure rule would allow only one real interpretation of (62) (ignoring that where each character is a separate symbol). Namely, its right-hand side consists of the morpheme string [anything + at + all]. Far from allowing free expression, the rule allows only a single output. To get what Jackendoff wanted, one needs minimally the equivalent of a logical variable over the class that E represents. But phrase structure rules and generative rules in general do not contain such equivalents. So, while Jackendoff (1984) insightfully uncovered and discussed a relevant class of evidence that showed that (1a, b, and c) were all untenable, unfortunately he did not then grasp the implications of what he had found.25 The relation between the false (1a) and (1b) is logically more complex than might appear. Certainly the falsehood of (1a) shows that (1b) cannot be true. Given that NLs are open and hence not recursively enumerable collections, the idea that one can only study them through finite grammars that generate (recursively enumerate) their member sentences is untenable. It cannot sensibly be required that one adopt descriptive mechanisms that must logically fail. Notably, no argument has ever been given by Chomsky (or anyone else as far as I know) for (1b), either. This is fundamental because in fact, once one goes beyond the uncontroversial issue of grammars being finite, even the truth of (1a) would not justify (1b). The reason is that there is nothing inherent in the linguistic goal of specifying the nature of infinite collections of highly structured objects like the sentences of NLs that imposes the methodology of formulating generative/constructive/prooftheoretic grammars. This is an obviously impossible conclusion given the existence of nonrecursively enumerable infinite collections, for example, the real numbers. Formal fields other than linguistics, mathematics, logic, and such, have thus devel-

200


oped ways to study such collections in various ways via appeal to axiomatic systems, model-theoretic satisfaction, and so forth. The point with specific reference to linguistics has recently been nicely put by Pullum and Scholz (2001: 1): (63)

“The second half of the 20th century saw the emergence of two quite different types of frameworks for theorizing about the syntax of natural languages. One sprang from the syntactic side of mathematical logic, the other from the semantic side.”

By the “syntactic side of mathematical logic” Pullum and Scholz refer to prooftheoretic (‘derivational’) approaches whose linguistic instantiations are the various versions of generative grammar. By the “semantic side of mathematical logic” Pullum and Scholz refer to model-theoretic/satisfaction approaches. About this they say (2001: 3): (64)

“It applies model theory rather than proof theory to natural language syntax.”

Given the logical existence of these two very different approaches, even if, counterfactually, (1a) were true and NLs were recursively enumerable collections, to justify generative grammar one would still need a never supplied argument that their proper grammars are generative/proof-theoretic systems, rather than nonconstructive, axiomatic/model-theoretic ones. Chomsky’s claim (1b) thus further begs the question of the superiority of proof-theoretic over model-theoretic approaches to NL grammars, which would still need to be argued even if (1a) were true. Since it is false, the missing argument can never be supplied. Since (1b) is false because (1a) is and (1b) would not follow even if (1a) were true, (1c) cannot, of course, be a tenable condition to impose on grammars. It simply embodies the question begging about the choice between proof-theoretic and modeltheoretic approaches to NL grammars. The weakest condition that can reasonably be placed on NL grammars is not the quite arbitrary (1c) but rather that they be systems capable of characterizing the full collections that form NLs. Given the openness of NLs and the arguments that the class of direct speech, geometric, and so forth, NL sentences are of the order of the real numbers, model-theoretic approaches seem to be the only basis for constructing correct NL grammars. The just expressed conclusion in effect reiterates one already reached in Langendoen and Postal (1984: 77–78), who said: (65)

“Since the ideas of generative grammar became dominant in the late 1950s, linguistics has in general assumed that the task of grammatical theory involves answering the question: What is the right form of generative grammar for NLs? The many disputes which have divided linguists over the past quarter century are then reducible by and large to disputes over claims about ‘right form’. Some linguists have believed that NL grammars contain transformational rules; others have denied this. Some linguists have believed that transformational rules are parochially ordered; others have denied this. Some linguists have believed that there are interpretive semantic rules; others have denied this. And so on. Underlying all such disputes has been the assumption that it is


201

possible through appeal to some combination of proof-theoretical devices to construct some generative grammar for each NL.”

Langendoen and Postal (1984) rejected this possibility but only reached the conclusion of the untenability of proof-theoretic grammars on the basis of the proof, mentioned at the outset, that NLs were not sets. Given the radical and controversial nature of that line of argument, this work has sought to construct a path to the same conclusion in a different and I hope less controversial way. This path appeals of course to the conclusion that NLs are not closed. Accepting the latter as a fact means that NLs cannot be regarded as recursively enumerable sets and directly yields the result that NL grammars cannot be generative grammars. The proof-theoretic approach to NL grammars, which generative grammar insists on, is thus seen as irreparably flawed independently of the bases of the argument in Langendoen and Postal (1984, 1985). The seemingly only available alternative then is to develop model-theoretic approaches, as primitively attempted in Johnson and Postal (1980) and Langendoen and Postal (1984: chapter 5.2). See Pullum and Scholz (2001) for many references to more recent sophisticated and formalized approaches to model-theoretic grammars. Finally, it is worth indicating exactly why a model-theoretic approach to NL grammars is not undermined by NL openness in the way that proof-theoretic approaches are. Consider the direct speech case (66): (66)

Maureen grunted ‘fnstribkl’.

A generative grammar cannot characterize such sentences because it cannot find (‘look up’) the alexical object constituent or its components in its lexicon. Thus no prooftheoretic grammar can, as required for factual adequacy, yield a proof (derivation) of arbitrary grammatical direct speech sentences simply because the unregimented direct speech constituents are unlistable, hence unlisted. But model-theoretic approaches, as stressed in Pullum and Scholz (2001), embody necessary (and, I would stress, sufficient) conditions for sentencehood. They state what conditions an object must satisfy to be a sentence of the NL described. Therefore, such a grammar can characterize (66) without having to mention fnstribkl or any other direct speech segment. For this to work, it merely must characterize the realizations of the object of direct speech verbs like grunt in such a way that no conditions are imposed that fnstribkl and so forth will fail to satisfy. A nonconstructive grammar can allow alexical constituents, regimented or not, in various contexts simply by failing to state overly specific constraints. For the object context of (66), it will suffice if the grammar requires only that an acceptable object constituent specify a set of actual noises. Thus a collection of endlessly many direct speech objects like that of (66) is characterized without mentioning any of them, engendering no conflict between a finite grammar and a nonrecursively enumerable set of constituents, that is, a class of constituents that have no gödel numbering; see note 9. Put differently, a model-theoretic approach permits accomplishing what Jackendoff’s putative but nonexistent phrase structure rule could not. So, while the openness of NLs is strongly incompatible with proof-theoretic approaches to NL grammars, hence with the defining ideas of generative grammar, it is straightforwardly compatible with model-theoretic approaches.


II

STUDIES OF JUNK LINGUISTICS


7

Junk Syntax 1 A Supposed Account of Strong Crossover Effects

1. Background The strong crossover phenomenon, apparently first treated in Postal (1971), designates binding failures like those in (1):1 (1)

a. b. c. d. e.

⊄Who1 did Frank convince her1 that you would hire t1? ⊄the principle1 which1 I inferred from it1 that no other principle entailed t1 ⊄What1 Jane compared it1 to a model of t1 was the Eiffel Tower. ⊄[Generalissimo Garcia]1, no one could persuade him1 that you were related to t1. ⊄It doesn’t matter [who]1 they claim she1 believes you should invite t1.

Following Wasow (1972, 1979), I refer to the asymmetric relation between antecedent and pronominal form, reflexive or not, as anaphoric linkage. Binding is thus a subtype of this. First noticed in 1968, examples like (1) manifested previously unknown restrictions on anaphoric linkages between extracted elements and pronouns. My original research subsumed these facts under the rubric ‘crossover phenomena’, a term taken to cover considerably more data, much of which subsequent work indicates is distinct from (1). Specifically, Postal (1971) failed to distinguish what Wasow (1972) I think properly differentiated as strong versus weak crossover binding violations, the former represented by (1), the latter by, for example, (2): (2)

a. ⊄Who1 did all of his1 associates detest t1? b. ⊄the proposal which1 your rejection of it1 led me to abandon t1 c. ⊄[Whatever starlet]1 they convinced her1 employer that you had interviewed t1, . . . 205

206


A major distinction is their different ranges of applicability. Some extractions induce no weak effects but do induce strong ones, for example, English topicalizations of definite referential NPs; see (3): (3)

a. Strong Crossover Case: ⊄Jenny1 I am sure she1 thinks you really dislike t1. b. Weak Crossover Case: ⊂Jenny1 I am sure her1 husband thinks you really dislike t1.

This English-internal difference correlates with the cross-linguistic fact that there do not even seem to be purported counterexamples to the strong effect; in contrast, weak effect variation exists even between French and English; see Postal (1993b).2 This study, which is restricted to strong violations, has two aspects. First, it is an extended rejection of the validity of the assertion in (4) that strong effects reduce to one of the elements of Chomsky’s binding theory: (4)

Chomsky (1981: 193) “Principle (C) gives the basic facts of strong crossover in the sense of Wasow 1972, 1979) . . .”

Referencing a notion of binding based on coindexing and c-command, Chomsky’s Principle C requires so-called R(eferring)-expressions to be unbound.3 Second, intertwined with the argument for the inadequacy of a reduction of strong effects to Principle C is an argument that the standards utilized in the work that supposedly supports claim (4) are so low as to qualify as junk syntax. Although Chomsky (1982b) transitorily abandoned the Principle C view of strong violations, it appears today to be standard and rarely challenged received wisdom about strong effects. Occasional alternatives like the proposal by Higginbotham (1980b) or that of Koopman and Sportiche (1982/1983) or the NP Structure proposals of Riemsdijk and Williams (1981, 1986) have few current echoes. Some sense of the scope of recent support for (4) is seen in (5): (5)

a. McCloskey (1990: 212) “Sentence (34b) is ungrammatical because the lexical NP John is c-commanded by, and coindexed with, the epithet the bastard. Sentence (34c) is ungrammatical because the embedded subject trace is A-bound by the epithet the bastard, giving rise to a Condition C (strong crossover) violation.” b. Georgopoulos (1991: 37) “Strong crossover (SCO) is an effect of principle C of the binding theory or an equivalent c-command condition, which prevents a variable or an r-expression from having an antecedent in a c-commanding A-position.” c. Ristad (1993: 85) “Recall that strong crossover is the configuration where an anaphoric element ccommands the trace of a displaced wh-phrase and intervenes between the wh-phrase and its trace as well.” d. Other Works That Invoke a Principle C Account of the Strong Crossover Effect Cinque (1990: 150), Cowper (1992: 170), Culicover (1997: 316, 325, 326), Fiengo and May (1994: 279–280), Fox (2000: 132, note 21), Haegeman (1991: 380), Harbert

JUNK SYNTAX

1

207

(1995: 182), Hornstein (1995: 21), Huang (1995: 139), Kennedy (1997: 702), Lasnik and Uriagereka (1988: 41; notably, though, this work, pages 137–138, expresses strong doubts about the Principle C account based on observations of Higginbotham), Müller (1995: 163), Napoli (1993: 487), Ouhalla (1994: 214–215), Roberts (1991: 17; 1997: 147), and Sells and Wasow (1999: 21).

Despite this broad acceptance, I argue that no Principle C account of the strong effect has ever been viable. The discussion is limited to English, which imposes no serious implications since at issue is the correctness of a claimed universal account. Therefore, a showing that it fails for English is entirely sufficient. The attempt to reduce the strong effect to Principle C violations is inevitably linked to independent proposals. These include notably those of (6): (6)

Key Assumptions Linked to a Principle C View of the Strong Effect a. The assumed ‘gaps’ in the relevant sentences, that is, cases of putative movement to so-called nonargument positions, are taken to be filled by objects called traces. Like all traces, these are bound by the assumed extractees in the relevant constructions. b. The postulated traces must be characterized as R-expressions, because that is the (only) category that Principle C restricts.

Assumptions (6a and b) each involve potential weaknesses, some of which, I claim, are genuine flaws. Even granting (6a and b) however, the specific factual requirement in (7) must be met: (7)

Each gap that induces a strong violation, a gap taken to be a trace under assumption (6a), must be c-commanded by the pronoun whose link to the gap position yields the violation.

Assumption (7) holds since the concept ‘binding’ Principle C appeals to is only instantiated by pairs of constituents in a c-command relation. The link to c-command also creates a potential weakness argued to be irreparable. The trace-based Principle C view of the strong effect grew out of Chomsky’s own earlier trace-theoretic description of the effect, in turn a development of the account in Wasow (1972), whose key idea was attributed to Peter Culicover. See (8) and (9): (8)

Wasow’s (1972, 1979: 160) number (10) a. <S1 He said <S2 Mary kissed someone S2> S1> b. < S1 Who did he say < S2 Mary kissed D S2> S1> c. Wasow (1972, 1979: 160) “The transformation of WH-fronting converts a structure like (10a) into one like (10b). Now, if who and he in (10b) are to be allowed to enter into an anaphoric relation, the Transitivity Condition requires that D and he also be anaphorically related. Consequently, the resultant sentence will be ungrammatical for the same reason that (11) is.” d. Wasow’s (11) was my *(10b).

208 (9)


a. Chomsky (1976, 1977a: 195) “Thus we can account for the full range of interpretations in (74) by appeal to independently motivated principles of anaphora, again on the assumption (72) that surface structure determines LF with the natural additional assumption that bound variables function (to first approximation) as names.” b. Chomsky (1976, 1977a: 195) “[N]or need we invoke any principle beyond established principles of anaphora that apply in (77)–(79).”

Space precludes discussion of these earlier approaches, but a crucial idea in all of them, as in the Principle C view, is that strong effects reduce to the anaphoric linkage restriction in nonextraction examples like (10): (10)

a. ⊄She1 convinced me that I should help Isabelle1. b. ⊄He1 said Mary kissed someone1.

Such a reduction played no role in the account of Postal (1971) and, much more importantly, is argued here to be incorrect. The traces invoked in the original Principle C proposal were so-called empty categories. More recently, its formulator has adopted the different account in (11), the so-called copy theory of traces: (11)

Chomsky (1995b: 202) “[T]he trace left behind is a copy of the moved element, deleted by a principle of the PF component in the case of overt movement. But at LF the copy remains, providing the materials for ‘reconstruction’.”

The two alternative views of traces yield two variants of the claim that strong effects follow from Principle C; these have different factual consequences and hence potentially distinct truth values. I argue that neither is tenable, beginning with the initial, empty category proposal.

2. Older objections to Principle C accounts 2.1. Remarks Despite its frequent invocation, at least four defects of the empty category version of the Principle C approach already appear in the literature. 2.2. Defect 1: stipulation of “R-expression” status Principle C can at best yield the strong effect only via the claim that the traces left by movements to nonargument positions are R-expressions. Unless this categorization follows from something, the degree of explanatory success achieved even if the strong effect could reduce in this way to Principle C would be less than claimed. Even sympathizers with Chomsky’s approach recognize this defect. Thus Higginbotham (1983:

JUNK SYNTAX

1

209

407) calls taking variables as R-expressions “rather unnatural” before giving cases where it does not work correctly (under a Principle C approach). Koster (1987: 68– 69) calls the assignment an “unnecessary stipulation.” And Lasnik and Uriagereka (1988: 42) defend the stipulation only by the faint praise that if all NPs have to be assigned to some category, “R-expression seems a not unreasonable candidate.” But the requirement does not follow since even if it is claimed that all movements leave traces, not all can be claimed to leave R-expression traces. This is impossible for so-called head movements and, of course, for movements to argument positions, whose traces must be anaphors. Very different consequences would obtain if movements to argument positions left R-expression traces and those to nonargument positions anaphors, both left anaphors, or both left R-expressions. So the actual choice from the four logical possibilities remains unprincipled. Moreover, Meyers (1994: 285) observes that subsuming only those traces linked to movement to nonargument positions and ordinary lexical NPs under a blanket category ‘R-expression’ is independently suspicious, for the former require antecedents, like the traces of movement to argument positions, but the latter do not. So the needed grouping links elements that contrast in required antecedence and fails to include others that do require antecedence. The original Principle C treatment then requires a dubious stipulation like (12): (12)

Traces of movement to nonargument positions are R-expressions.

2.3. Defect 2: non-NP extractions A second problem for the original Principle C account also relates to dependence on (12). For this is too general, since ‘R-expression’ is only a category of NPs, while non-NPs are also taken to move to nonargument positions. Chomsky’s binding theory that includes Principle C is exclusively an account of certain NP properties. Incidentally, the recent invocation of DPs instead of NPs has no relevant consequences here and will here be ignored. Hence (12) has to be replaced, by something like (13): (13)

(Only) traces of movement of NPs to nonargument positions are R-expressions.

Therefore, a claim that the strong effect reduces to Principle C entails that such effects are never induced by non-NP extraction, specifically not by Prepositional Phrase (PP) extraction. This follows since Chomskyan traces of moved PPs must be PPs, hence not ‘R-expressions’. If the trace of a moved PP were an NP, a violation of Chomsky’s (1981) projection principle would ensue. As Lasnik and Uriagereka (1988: 41) put it: “In accord with Trace Theory, let us assume that who leaves a trace when it moves. This trace is obviously an NP, a fact ensured by Trace Theory, which essentially says that, upon movement, an item leaves behind a syntactic silent copy of itself.” But the entailment is wrong. For as Koster (1987: 82) indicates: If a PP containing a Wh-word is preposed, we have a really crucial example: (130) *[With whomi] did hei say that Mary talked [PP t ] This is a normal case of strong crossover. It is not possible to construct a reading in which there is a binding relation between he and (the variable corresponding

210


to) whom. In this case, the example is not ruled out by the binding theory, because the binding theory says nothing about PP-traces.4

While I will not in general quarrel with Principle C itself, it has known inadequacies and a possibly not so well known one. Namely, since Chomsky’s binding theory is restricted not only to argument positions but also to NPs, an evident difficulty is that the bad CP anaphora case in (14a) is unblocked although the no more impossible gerundive NP version is properly blocked: (14)

⊄It1suggests that Mike thought a. [that 2 and 2 is 4]1/b. about [2 and 2 being 4]1.

Example (14a) satisfies Principle C since the that clause, a CP, is thereby not an R-expression. The lack of induced binding failure in (14a) leads to a further nonNP related crossover problem linked to examples like (15a): (15)

a. ⊄[That Ted is a spy]1, I now realize that it1 indicates that Bob knew t1. b. *Ted adores/condoned/criticized/studied/ that Marsha is a vampire. c. *[That Marsha is a vampire]1, Ted adores/condoned/criticized/studied t1.

Such CP topicalizations are in general only licit when the t1 position otherwise accepts that clauses, as (15b and c) illustrate, which suggests (15a) represents that clause topicalization. But via the same logic by which Principle C fails to block of (14a), it would also wrongly not yield a Principle C violation in the strong effect case (15a), either. 2.4. Defect 3: the secondary strong crossover effect A third difficulty with the original Principle C claim relates to data like (16), also found in Postal (1971: 90): (16)

a. [Whose1 cousin]2 did you convince ⊄him1/⊄him2 I had run over t2? b. the nurse [[whose1 father’s]2 sisters]3 I convinced ⊄her1/⊄him2/⊄them3 that you would contact t3

Here the trace Chomsky’s analysis posits, even if categorized as an R-expression and c-commanded by a pronoun, is only wrongly bound under Principle C when linked to the entire questioned or relativized phrase. But anaphoric linkage is equally banned for the contained NPs, which do not link to any traces. Further terminology is helpful. I take restrictions that ban anaphoric linkages like that between pronouns and the entire extracted phrases in (1) and (16) to represent the Primary strong effect and those that ban such linkages between the pronouns and (certain) subconstituents of the entire extracted phrase to constitute the Secondary strong effect. The latter has been extensively discussed; see, for example, Barss (1986), Chomsky (1981: 89), Culicover (1997), Engdahl (1986: 302), Higginbotham (1980a, 1980b, 1983), Jacobson (1976, 1977), Koopman and Sportiche (1982/1983), Koster (1987: 81), Kuno (1987: 55f, 60), Lasnik and Uriagereka (1988), Riemsdijk (1982), Riemsdijk and Williams (1981,

JUNK SYNTAX

1

211

1986), and Safir (1996, 1998). But there seems to be no firmly established tracetheoretic solution. And as Kuno (1987: 61) asserts: “Chomsky [(1981)] admits that there are many problems that arise in this connection but says that it is beyond the scope of his book. He dismisses trace theory with structured trace and continues to adopt trace theory with empty category. He thus leaves unexplained the fact that he cannot be coreferential with whose in (8.14).” Kuno’s (8.14) is an example like (16a). Minimally then, Principle C applied to the Surface Structure, which is what was originally claimed,5 fails to induce all the attested strong effects. So the secondary strong effect undermines the original idea that recognition of traces could reduce the strong effect to a c-command condition that references Surface Structures.6 Much discussion of the secondary effect involves appeal to so-called reconstruction, which time precludes discussing. But I return to the issue in effect when discussing the copy trace view. 2.5. Defect 4: the asymmetry property The gravest previously noted defect of Principle C accounts of the strong effect relates to an asymmetry. In standard strong effect cases like (1), the extractee is the antecedent of the pronoun whose ‘crossing’ yields the violation. It is such cases, which involve question phrases or relative pronouns, that Culicover, Wasow, and Chomsky concentrated on almost exclusively; for them, a Principle C approach might have some initial plausibility. But consider (17): (17)

a. Postal (1971: 143): ⊂Myself1, I1 can’t begin to understand t1. b. Postal (1971: 143): ⊂[To myself1]2, I1 never send things t2.

Evidently, early strong effect work considered anaphoric linkages where a topicalized reflexive pronoun or PP that contained such was well formed in typical strong effect contexts. Chapter 16B of Postal (1971) was devoted to the fact that pronouns could licitly topicalize over their antecedents. Special devices were proposed to deal with the contrast between these cases and (18): (18)

a. ⊄Fred1, he1 can’t begin to understand t1. b. ⊄[To Fran1]2, she1 never sends things t2.

Yet under Chomsky’s empty category Principle C account the well-formed binding in (17a) yields no less a violation of Principle C than the ill-formed binding of (18a), as Kuno (1987: 81) already observed: “But trace theory does not distinguish . . . the trace of a full NP and that of a pronominal or reflexive, and therefore is incapable of distinguishing the two situations. This casts a serious doubt on the very foundation of trace theory.” The same factual pattern emerges in cleft structures and the restriction to extraction to the front of the clause immediately containing the antecedent characteristic of (17) is arbitrary:

212 (19)


a. b. c. d.

⊂Herself1, I am sure that Gladys1 doesn’t want to vote for t1. ⊂It is herself1 that I am sure that Gladys1 doesn’t want to vote for ⊂Himself1, they found out that Eddy1 had talked to Edna about t1. ⊂It was himself1 that they found out that Eddy1 had talked to Edna about t1.

Postal (1971) gave one well-formed example, namely, (20), which contained an extracted nonreflexive pronominal: (20)

Postal (1971: 158): ⊂[To him1]2, [the man]1 claimed you were engaged t2.

But that work took (20) to be exceptional and assumed topicalization of a nonreflexive pronoun in standard strong effect configurations yielded violations; see (21): (21)

Postal (1971: 143) (see also Postal [1971: 145, 149]) a. ⊄Her1,Barbara1 claimed that Tony hated t1. b. ⊄Him1, Harold1 wanted Betty to visit t1. c. Postal (1971: 158): ⊄Him1, the man1 claimed you were engaged to t1.

In retrospect, my earlier claims about (21) seem fundamentally mistaken. While certain speakers reject such examples,7 they do not present the sharply impossible anaphoric linkages of true strong effect violations. Topicalized pronoun examples of this sort should thus not be assimilated to sentences obtained by interchanging the relevant antecedents and pronouns, which yields unchallenged strong effect violations like (22):8 (22)

a. b. c. d.

⊄Tony1, he1 said Harry insulted t1. ⊄[That man]1, he1 claimed you were engaged to t1. ⊄It was Tony who1 he1 said Harry insulted t1. ⊄It was that man whom1 he1 claimed you were engaged to t1.

Summarizing then, the strong effect is limited as in (23): (23)

Strong effects occur only in structures in which: a. There exist phrases A, B such that A antecedes B; b. A is extracted.

While (23a) restates a banality, (23b) distinguishes cases like (22) from those where a pronominal is extracted, as in (21) and (i) of note 8. I refer to this difference as the Asymmetry Property. A viable account of the strong effect must deal with minimal Asymmetry Property contrasts like (24): (24)

a. b. c. d.

⊄Who1 did the directors convince him1 that Jane should vote for t1? ⊂It was himself that1 the directors convinced Jane that he1 should vote for t1. ⊄the dancer who1 I promised her1 that you would visit t1 ⊂Her(self)1, Joan1 never promised me that you could visit t1.

JUNK SYNTAX

1

213

But Chomsky’s original Principle C account draws no such distinction and so is again robustly counterexemplified by data easily available in Postal (1971). The work cited in (4) simply ignored these facts.9 More generally, the asymmetry problem was apparently not recognized internal to published trace-theory literature until fairly late in the 1980s, although it is noted in earlier unpublished theses; see Barss (1986), Browning (1987), and Sportiche (1983). For published trace-theoretic work, the earliest citations seem to be Williams (1986: 288) and Koster (1987: 78–79). The latter remarks: If there is a construal chain that makes it possible to identify the trace with reflexive features, the sentence is grammatical, even if the trace is A'-bound from COMP: (122) Himselfi [ Oi [hei does not really like ti ]]

Lasnik and Uriagereka (1988: 81–82, 157) consider material parallel to that in Kuno (1987: 81–82), showing the Principle C account makes erroneous predictions about topicalization of reflexive and nonreflexive pronouns. But this work treats the issue as newly noted, ignoring the extensive support for this conclusion in Postal (1971).10 2.6. Summary So far then I have cataloged those objections to Chomsky’s original Principle C approach to the strong effect already found in the literature and listed in (25): (25)

a. the stipulative character of the critical assignment of relevant traces to the category R-expression; b. the existence of strong crossover effects in non-NP extraction cases; c. the secondary strong crossover effect; and d. most seriously of all, the Asymmetry Property.

3. New objections to Principle C accounts 3.1. Remarks Flaws in the original Principle C approach not to my knowledge already found in the literature include diverse cases where the c-command condition required for Principle C relevance fails to hold between offending pronoun and extraction site. 3.2. Defect 5: the offending pronoun is in a prepositional phrase The first such case is relatively minor. Example (26) shows that at best, c-command is not the right notion for any Principle C formulation intended to capture strong effects: (26)

a. ⊄[Which lawyer]1 did Mike fail to mention [PP to her1] that we had praised t1? b. ⊄[Which lawyer]1 did Mike fail to tell her1 that we had praised t1?

214


Since the pronoun in (26a) fails to c-command the trace position, no Principle C violation can be induced. Nonetheless, anaphoric linkage is as impossible as in (26b). This problem parallels others PPs raise for anaphoric description and might just support replacing c-command with the notion suggested for related reasons in (27): (27)

Pesetsky (1995: 173) “For example, we might define a relation called EBPP-command (‘everything-butPP-command’) as a component of the notion ‘binding’ (cf. Lasnik’s [1976] notion of Kommand). (453)

a. a EBPP-commands b if and only if the first non-PP that dominates a also dominates b. b. a binds b if and only if a EBPP-commands b, a precedes b, and a and b are coindexed.

“Such a theory would account for the interaction of prepositions with binding phenomena. Nonetheless, (453) would not explain this interaction. (453) cannot tell us why PPs fail to count for command, precisely because this fact is stipulated.”

A restatement of Principle C in terms of (27) would correctly block (26) and is independently motivated for Principle C by noncrossover data like (28): (28)

a. ⊄Mike failed to mention [PP to [some lawyer]1] that he respected [that lawyer]1. b. ⊂Mike talked [PP to Sharon1] about herself1.

So (26) indicates further sloppiness in earlier proposals that the strong effect reduces to Principle C but not a grave technical problem. 3.3. Defect 6: the offending pronoun is in a coordinate phrase The issue for a Principle C account raised by PP data has a partial analog in coordination data like (29), which represents the second new c-command problem. (29)

a. ⊄[Which nurse]1 did Mike convince Jim and her1 that you voted for t1. b. ⊄It was that nurse whom1 Jim and she1 said that you would hire t1. c. ⊄It doesn’t matter [what nurse]1 they arranged for Jim and her1 to tell you that I would hire t1. d. ⊄[What woman]1 did they place an alligator between Jim and her1 while interviewing t1?

These examples illustrate what I believe is the general truth in (30): (30)

If a context C that induces strong violations is mapped into a different one solely by replacing the offending pronoun P in C by a conjunction of NPs that include P, binding of P by the extractee remains impossible.

JUNK SYNTAX

1

215

Evidently, though, given that the pronoun in such coordinate structures is a subconstituent of a larger NP, it cannot c-command the trace position, and PP invisibility would not help. Designing some analog to (27) that ignores coordinate nodes would be misguided, given (31): (31)

a. b. c. d.

⊂Mike convinced Jim and her1 that you voted for [that nurse]1. ⊂Jim and she1 apparently (both) said that you would hire [that nurse]1. ⊂They arranged for Jim and her1 to tell you that I would hire [that nurse]1. ⊂They placed an alligator between Jim and her1 while interviewing [that woman]1.

These data show informally that undoing extractions like those in (29) yields examples that do not contain anaphoric linkage violations. So, contrary to instances considered so far, where violations in putative strong effect cases parallel those found between pronoun positions and premovement extractee positions, not so in (29)/(31). With respect to this and other cases introduced later, a Principle C defender might deny that the problematic cases, here (29), represent strong effects. Such an approach is suspect because it means recognizing some independent binding constraint to block (29). Worse, examples like (29) manifest key defining features of the strong effect; they are insensitive to the PP/NP distinction, manifest the Asymmetry Property, and reveal primary and secondary variants; see (32): (32)

a. b. c. d.

⊄Jane1, [from whom1]2 Ted and she1 said the police had hidden the truth t2, ⊄Jane, who(m)1 Ted mentioned to Mike and her1 that you would call t1. ⊂It was her who(m)1 Mike and Jane1 said you would call t1. ⊄[Whose1 niece]2 did Mike and ⊄she1/⊄she2 claim that you had insulted t2?

The Principle C account’s failure to handle coordination facts is thus hardly to be rectified by invoking a phenomenon distinct from the strong effect. While the coordinate data just cited raise real issues for a Principle C account, these are relatively tame when compared to facts like (33): (33)

a. ⊄It was [Jane1 and Barbara2]1,2 who1,2 Mike respectively convinced her1 that you would call t1and tried to convince her2 that I would call t2. b. ⊄[Which two women]1,2 did Mike respectively convince her1 that you would call t1 and try to convince her2 that I would call t2?

Here a Principle C treatment based on traces not only fails to induce the manifested strong effects. Rather, movement/trace approaches to extraction have not even shown how to provide basic analyses of such structures and, of course, standard treatments of extraction in general do not even mention them. 3.4. Defect 7: the offending pronoun is in an exceptive structure The third new objection connected to a failure of c-command between pronoun and extraction site relates to exceptive structures such as (34):

216 (34)


a. ⊄[Nobody but/except (for)/other than Vanessa1]2 could they convince her1 you would invite t2. b. ⊂[Nobody related to/fond of/interested in Vanessa1]2 could they convince her1 you would invite t2. c. ⊂[Nobody but her1]2 could they convince Vanessa1 you would invite t2. d. ⊄Vanessa1, they convinced [everybody but her1]2 that you would invite t1. e. ⊂Her1, they convinced [everybody but Vanessa1]2 that you would invite t1.

While it might be claimed that (34a) is not really a new problem, since it represents essentially only a further type of secondary strong effect, this is really wrong, for normally there is no secondary strong effect with definite referential NPs, as the contrasting (34b) shows. With respect to this property, there may, however, be individual variation; for instance, the anaphoric linkage in (35) is marked bad, while for me it is perfect: (35)

Pesetsky (1995: 270) ⊂/⊄[Which picture of Tom1]2 did he1 say Sue had purchased t2?

Example (34c) is not incompatible with a Principle C account, as that condition is satisfied. But a further genuine problem is seen in (34d). This does manifest illicit anaphoric linkage, although the pronoun cannot c-command the extraction site; so Principle C is not invoked. That (34a and d) are nonetheless strong effects is argued by the well-formed anaphoric linkages in (34c and e), which show that anaphoric linkages here manifest the Asymmetry Property. A Principle C account defender might invoke here some analog of the concept proposed by Pesetsky or, alternatively, might claim that the index associated with the inner NP in an exceptive structure somehow percolates up to the outer NP. This could yield Principle C violations by converting (34a) to (36a) and (34d) to (36b): (36)

a. ⊄[Nobody except Vanessa1]2/1 could they convince her1 you would invite t2. b. ⊄Vanessa1, they convinced [nobody but her1]2/1 that you would invite t1.

In (36), the traces are arguably illicitly bound under Principle C by coindexed nodes. Precedents for such moves appear in trace-theoretic work on anaphora, for example, Safir (1984, 1996), who proposes such a mechanism for structures like those involved in secondary strong effects; see (37): (37)

Safir (1996: 325) “(31) a. Q-chain A Q-chain is a sequence of adjacent A-binding constituents [O1, O2, . . . ,On] such that Om-1 binds a variable in Om. The initial O1 of the Q-chain is the Qchain head. The variable bound by 0n is the Q-chain variable. The Q-Chain Convention Add the index of the Q-chain head to that of the Q-chain variable. The extractions that create and extend Q-chains are predicated on the LF movement of the class of elements that move scopally at LF, namely, the socalled true quantifiers.”

JUNK SYNTAX

1

217

But the particular mechanism Safir advanced could not function for exceptives, as it only creates secondary index assignment that originates from quantificational expressions, not with definite referential ones like Vanessa in (34a). Moreover, claiming that the inner index of an exceptive is in general obligatorily assigned to the outer NP is untenable, given (38b, c, and d): (38)

a. b. c. d.

⊄[Nobody ⊂[Nobody ⊂[Nobody ⊄[Nobody

except except except except

Vanessa1]2/1 praised her1. ??her1/herself1]2/1 praised Vanessa1. Vanessa1]2/1 praised Vanessa1. Vanessa1]2/1 praised herself1.

Obligatory secondary assignment would be not only consistent with (38a) but also supported by it, given that it would block the bad anaphoric linkage under anything like Chomsky’s binding Principle B. But it is nonetheless counterexemplified by (38b and c), where it would wrongly create Principle C violations, and by (38d), which it would wrongly claim satisfies Principle A. It is not directly relevant here that the reflexive variant of (38b) creates a problem for any analog of Chomsky’s Principle A, as there is no obvious way the apparent anaphor herself could be bound. Thus even if appeal to index reassignment mechanisms is allowed, it offers no way to keep Principle C consistent with data like (34a and d), under the empty category view of traces. Facts rather parallel to those with exceptives are found with phrases constructed with only and even; see for example (39): (39)

a. b. c. d.

⊄Vanessa1, they convinced [even/only her1]2 that you would invite t1. ⊄Vanessa, who1, they convinced [even/only her1]2 that you would invite t1, . . . ⊂Vanessa1, they convinced [her1 mother]2 that you would invite t1. ⊂Vanessa, who1, they convinced [her1 mother]2 that you would invite t1, . . .

Contrasts like that between (39a, c) and (39b and d) argue against treating (39a and b) as weak crossover violations. Space forces me to pass over the rather parallel ordinary quantifier structures in (40) and (41): ⊄[Each one/All/None of [those starlets]1]2 praised them1. ⊄[None of [those stars]1]2 could they convince them1 that you would invite t2. ?⊄[None of [them1]]2 could they convince [those stars]1 that you would invite t2. ?⊄[Those stars]1, they could convince [none of [them]1]2 that you would invite t1. ⊂Them1, they convinced [none of [those stars]1]2 that you would invite t1.

(40)

a. b. c. d. e.

(41)

⊄[All of them1/*themselves1]2/1 praised [those stars]1.

3.5. Defect 8: the noncrossover effect A fourth largely new objection to a Principle C treatment linked to its reliance on c-command involves data that lacks a property invariably present in all the strong ef-

218


fects cited so far. In previous data, the extraction site uniformly follows the pronoun that the extractee cannot antecede. This property is not relevant to the Principle C account, but it was, of course, to the proposal of Postal (1971), where it determined choice of the term ‘crossover’. Nonetheless, (42) exemplifies cases that manifest the key features of standard strong effects other than the word order property whose extraction sites nonetheless precede the pronouns with which anaphoric linkage is impossible: (42)

a. Jacobson (1976: 12) attributed to William Leben (i) *⊄Who1 did the wolf mention his planning to eat t1 to her1? (ii) *⊄Who1 did Hamlet talk about his overhearing t1 to him1? b. Koopman and Sportiche (1982/1983: 149) and Safir (1984 :605) ⊄Who1 did you give a picture of t1 to him1? c. ⊄Who1 did you paste photos of t1 on him1? d. ⊄the nurse who1 they bought sketches of t1 from her1 e. ⊄Jerome1, Ira expressed contempt for t1 near him1. f. ⊄It was Jerome1 that Ira expressed contempt for t1 near him1.

While (42a–f) resemble standard primary strong effect violations in having extracted NPs that cannot anaphorically link to certain pronouns, they contrast in another respect. As in the coordinate cases, the corresponding pre-extraction structures violate no constraint on anaphoric linkages and so not Principle C; see (43): (43)

a. ⊂You gave a picture of Claude1/[some officer]1/someone1 to him1. b. ⊂They bought sketches of [some/that nurse]1 from her1. c. Koopman and Sportiche (1982/1983: 148) explicitly note the lack of c-command between the antecedent and pronoun positions in cases like (43a and b).

Actually, some speakers reject the indefinite versions of (43). I find them at worst stylistically heavy. Moreover, anaphoric linkage is less problematic when the pronouns are replaced by anaphoric lexical NPs; see (44): (44)

⊂Earl gave a picture of [some officer]1 to [that officer]1.

This is relevant because such anaphora is also subject to strong effects. I refer to the anaphoric linkage violations revealed in (42) as noncrossover effects. Some might hope to reduce noncrossover effects to the c-command requirement of Principle C via special constituent structure assumptions. Given that the linguistics from which the Principle C account arose is given to a vast expansion of recognized constituents, for example, functional projections that include agreement constituents, multiple constituents with empty heads, recursions on nodes like VP, and so forth, the possibilities are not small. Nonetheless, good evidence that such moves, whatever their independent validity, cannot succeed in reducing noncrossover effects to c-command conditions is provided by the fact that the extraction position can even be inside a subject. The relevance of this requires a brief codicil. Even granting that, as generally claimed, subjects are islands, which in general bar extraction, and ignoring marginal

JUNK SYNTAX

1

219

acceptability cases like (45), one can still use such ungrammatical extractions to test binding hypotheses because of principle (46): (45)

Reinhart (1983: 120) ?[Which businessman]1 did the gossip about t1 cause a national scandal?

(46)

Mere extraction from an island, even when yielding severe ill-formedness, does not inherently block anaphoric linkages if such are licit in the pre-extraction structure itself.

The truth of (46) appears in the fact that while (47b, d, and f) are sharply ungrammatical, there is no more interference with the indicated anaphoric linkages than there is in the nonextraction cases (47a, c, and e): (47)

a. b. c. d. e. f.

⊂I found that Jane and Mark1 both said you would hire him1. ⊂*It was Mark who1 I found that Jane and t1 both said you would hire him1. ⊂I compared nobody but Michelle1 to the woman who hated her1. ⊂*It was Michelle who1 I compared nobody but t1 to the woman who hated her1. ⊂Because Carla1 was surly, Mike wouldn’t call her1. ⊂*It was Carla who1 because t1 was surly, Mike wouldn’t call her1.

Returning to noncrossover effects, regardless of the status of extraction from the subject, anaphoric linkage is impossible in (48a, b), (49b), and (50b), instantiating the noncrossover effect with a subject-internal extraction site. (48)

a. ⊄*[Which businessman]1 did gossip about t1 annoy him1? b. ⊄*a businessman1 who1 gossip about t1 was infuriating to him1

(49)

a. ⊂[Several friends of Jerome’s1] are talking about him1. b. ⊄*Jerome1, who1 several friends of t1 are talking about him1,

(50)

a. ⊂[No future teacher]1 did several friends of Vanessa’s2 describe t1 to her2. b. ⊄*It was Vanessa who2 [no future teacher]1 did several friends of t1 describe t1 to her2.

No tenable constituency assumptions could reduce these antecedence blockages to Principle C, because a subconstituent of a subject cannot c-command elements of the VP even under the wildest proliferation of nontraditional categories. Space also forces ignoring (51): (51)

a. ⊂[Most articles about Mary1]2, I am sure she1 hates t2. b. ⊄*It is Mary who1 [most articles about t1]2, I am sure she1 hates t2.

Reinforcing the point that strong effects are found in environments that lack c-command between a pronoun and its potential antecedent, (52) illustrates that the exceptive and noncrossover cases can be combined, with the expected strong effects:

220 (52)


a. ⊄[No professor except Marsha1]2 did they hand [pictures of t2] to her1. b. ⊄It was Marsha who1 they handed [pictures of t1] to no one but her1.

For noncrossover effects then, the idea that anaphoric linkage constraints associated with extractions reduce to constraints holding in pre-extraction structures, which has been the base of all trace-theoretic approaches to the strong effect, seems entirely untenable, which suggests that the violations are in some way induced by extractions themselves. That traditional strong effects and what are here called noncrossover effects are special cases of one unified phenomenon is supported by the existence of the four clear parallels between them listed in (53); see (54)–(66). (53)

Similarities between Standard Strong Effects and Noncrossover Effects a. Both effects have primary and secondary variants. b. Both types of secondary effect fail to appear when the extractees are, for example, definite referential NPs. c. Both effects also exist in cases of anaphorically linked lexical NPs instead of pronouns, a property noted for the standard strong effect in McCloskey (1990). d. Both effects manifest the Asymmetry Property.

First, there are restrictions that parallel those in secondary strong effect examples like (16), which leads directly to a distinction between primary and secondary noncrossover effects. As far as I can determine, the latter have the same general properties as secondary strong effects. So, alongside the primary noncrossover effect example (54a) are secondary noncrossover violations like (54b and c). See also (55): (54)

a. ⊄[Which man]1 did you paste photos of t1 on him1? b. ⊄[[Which man]1’s dog]2 did you paste photos of t2 on him1? c. ⊄the man [[whose 1 dog’s] 2 trainer] 3 you pasted photos of t3 on ⊄him1/⊄him 2/ ⊄him 3

(55)

a. Secondary Strong Effect: Higginbotham (1983: 407) ⊄[Which biography of [which artist]2]1 do you think he2 wants to read1? b. Secondary Noncrossover Effect ⊄[Which biography of [which artist]2]1 do you think I should show a review of t1 to him2?

Secondary noncrossover effect cases also parallel extraction from subject primary noncrossover effects such as (48b) and (49b); see (56): (56)

a. ⊄a businessman [whose1 son]2 gossip about t2 was infuriating to him1 b. ⊄Jerome1, [whose1 sister]2 several friends of t2 are talking to him1,

A second parallel between the two secondary effects strengthens their suggested unity. As touched on earlier, unlike question and restrictive relativization, certain

JUNK SYNTAX

1

221

extractions, topicalization, clefting, and some nonrestrictive relatives may fail to yield secondary strong crossover effects, as in (57). These same extractions also fail to induce secondary noncrossover effects under analogous circumstances; see (58): (57)

a. ⊂[Carl’s1 neighbor]2, they did not introduce him1 to t2. b. ⊂It was [Carl’s1 neighbor]2 that they introduced him1 to t2. c. ⊂Carl, [whose1 nurse]2, they did not introduce him1 to t2,

(58)

a. ⊂[Carl’s1 neighbor]2, they did not paste any pictures of t2 on him1. b. ⊂It was [Carl’s1 neighbor]2 that they pasted some pictures of t2 on him1. c. ⊂Carl, [whose1 neighbor]2, they did not paste any pictures of t2 on him1,

Consider (59a, b, c), where lexical NPs can be anaphorically linked: (59)

a. ⊂He gave a picture of a troll1 to that troll1. b. ⊂He pasted a picture of a troll1 on that troll1. c. ⊂Never show unflattering pictures of any trolls1 to [those trolls]1.

These are consistent with Chomsky’s binding theory since the NP pairs satisfy Principle C, the only constraint relevant to anaphoric linkage between lexical NPs. That is, since in (59) in none of the lexical NP pairs does one c-command the other, neither binds the other. This contrasts with the situation in, for example, (60): (60)

⊄A troll1 (said he1 had) pasted a picture of Rhonda on [that troll]1.

But in extraction correspondents of the well-formed (59b), such as (61), there is, as for noncrossover effects with pronouns, an effect not reducible to properties of preextraction structures: (61)

a. ⊄[Which troll]1 did he paste a picture of t1 on [that troll]1? b. ⊄List all the trolls1 that1 he pasted pictures of t1 on [those trolls]1.

For any traces inside the picture NPs in (61) neither bind nor are bound by the demonstrative NPs. So an analog of the primary noncrossover effect exists for anaphoric lexical NPs. Principle C would again at best have to be supplemented by some further principle. Riemsdijk and Williams’s treatment also fails for (61) since application of the non-c-command constraint on the NP-structures of the examples would not block them. That the primary noncrossover effect exists with nonpronouns provides a third parallelism with the primary strong effect. For as McCloskey (1990) observed, the latter also has a nonpronominal analog; see (62). Moreover, variants of both secondary effects also exist with lexical NPs; see (63): (62)

a. ⊄[Which nurse]1 did you convince that nurse1 that you would hire t,? b. ⊄a troll, [which troll]1 I saw [that troll]1 pretend t1 was an elf,

222 (63)


a. ⊄[Which biography of [which space alien]2]1 did that space alien2 prove t1 was slanderous? b. ⊄[Which biography of [which space alien]2]1 did Ernest write a letter about t1 to that space alien2?

The conclusion that the noncrossover and strong effects reflect the same principles is supported in a fourth way by the Asymmetry Property, for a similar feature arguably holds for the noncrossover effect. Compare (64a, b): (64)

a. ⊄It was Laura1 whom1 I described several photos of t1 to her1? b. ⊂It was her(self)1 whom1 I described several photos of t1 to Laura1.

(65)

⊂I described several photos of her(self)1, to Laura1.

(66)

a. ⊄It was Laura1 whom1 I persuaded her1 that you might hire t1. b. ⊂It was her(self)1 whom1 I persuaded Laura1 that you might hire t1.

The anaphoric linkage in (64b) lacks the impossible status of that of (64a), typical of strong effect violations, and seems as good as the nonextraction case (65). Significantly, the properties of (64) seem to be essentially identical to those of analogs in classical strong effect environments like (66). So the noncrossover effect also only exists when antecedents are extracted, as in (23). The extensive paradigmatic similarities between strong effects and noncrossover effects support taking both to reflect the same underlying principles. Since these cannot be reduced to Principle C for the strong noncrossover effect, it follows that this is incorrect for strong crossover effects as well. Chomsky considered a single noncrossover effect in passing as in (67), stating that (67ai and ii) were weak crossover violations: (67)

a. Chomsky (1982b: 38) echoing Koopman and Sportiche (1982/1983: 149) (i) ⊄Who1 did you give a picture of t1 to him1? (ii) ⊄Who1 did you give a picture of him1 to t1 ? (iii) “Both examples violate weak crossover as determined by the Bijection Principle.” b. ⊄Anthony1, I gave a picture of t1 to him1. c. ⊄It was Anthony who1 I gave a picture of t1 to him1. d. ⊄Anthony, who1 I gave a picture of t1 to him1, . . . e. ⊄Anthony1 was hard to give pictures of t1 to him1.

But this claim is undermined by the fact that (67b, c, d, and e), which also involve illicit anaphoric linkages, have definite referential extractees. For as noted at the outset and discussed in the works in (68a), such extractees do not in general induce weak crossover effects, as illustrated by the lack of effect in (68b, c, d): (68)

a. b. c. d.

Lasnik and Stowell (1991), Postal (1993b), and Safir (1996) ⊂It was Anthony whom1 his1 boss displeased t1. ⊂Anthony, whom1 his1 boss displeased t1, . . . ⊂Anthony1 was hard to describe his1 new office to t1.

JUNK SYNTAX

1

223

While seemingly favoring Chomsky’s claim, the badness of the anaphoric linkage in the definite extractee correspondent of (67aii) seen in (69a) is, I suggest, irrelevant. For I would argue that this anaphoric linkage violation is not a weak crossover effect but only a reflection of the requirement, whatever its basis, for a reflexive form in this environment; the presence of the latter yields the fine (69b): (69)

a. ⊄Anthony, who1 I gave a picture of him1 to t1, . . . b. ⊂Anthony, who1 I gave a picture of himself1 to t1, . . .

Example (69b) violates the Bijection Principle no less than (69a), indicating contra Koopman, Sportiche, and Chomsky the irrelevance of that condition to the facts under discussion. 3.6. Defect 9: extraction from adjuncts A fifth objection to Principle C accounts based on empty category traces connected to a failure of c-command involves extraction from adjuncts. Since this is often barred, principle (46) is again potentially crucial. As is well known, backward linking of a pronoun object to an antecedent in an adjunct is frequently permitted, as in (70c, e): (70)

a. b. c. d. e. f.

⊄The doctor told her1 that you loved Gladys1. ⊄It was Gladys who1 the doctor told her1 that you loved t1. ⊂The doctor told her1 that story while treating Gladys1. ⊄It was Gladys who1 the doctor told her1 that story while treating t1. ⊂It was her who1 the doctor told Gladys1 that story while treating t1. [Which patient’s1 child]2 did the doctor tell ⊄her1/⊄her2 that story while treating t2?

Here (70a) is an ordinary Principle C violation given that the object c-commands everything in the complement clause; (70b) is then a standard strong effect case attributed to Principle C, seemingly reducing to the same principle as (70a). Example (70c) is of course generally taken to show that an object does not c-command an element in an adjunct like that in (70c). If not, it is completely unexpected under a Principle C account that extraction from the position of Gladys in (70c) would yield a strong effect. But (70d) shows that it does. The contrast between (70d and e) reveals the Asymmetry Property, arguing that (70d) is a genuine strong effect, as does the secondary strong effect in (70f). Parallel facts are seen in the different adjunct cases in (71) and (72): ⊂The doctor jumped up enraged at her1 after arguing with Gladys1. ⊄It was Gladys who1 the doctor jumped up enraged at her1 after arguing with t1. ⊂?It was her who1 the doctor jumped up enraged at Gladys1 after arguing with t1. [Which patient’s1 child]2 did the doctor jump up enraged at ⊄her1/⊄her2 after arguing with t2?

(71)

a b. c. d.

(72)

a. ⊂The suggestion was never made to him1 that you might consult [Dr. Felix]1. b. *⊄[Dr. Felix]1, the suggestion was never made to him1 that you might consult t1.

224


c. *⊂Him1, the suggestion was never made to [Dr. Felix]1 that you might consult t1. d. *[Which patient’s1 child]2 was the suggestion made to ⊄her1/⊄her2 that you could play with t2?

The problem these cases raise is that to allow the anaphoric linkages in (70a), (71a), and (72a), the object pronoun position must not c-command into the adjunct. But to reduce the anaphoric blockages in, for example, (70b), (71b), and (72b) to Principle C, the object pronoun position must c-command into the adjunct. Such results are not jointly possible, since the positions in the corresponding good and bad cases are seemingly identical. These adjunct cases would, though, not be problematic if the suggestion of Williams given in full in (73) were valid. This claims that extraction from such adjuncts is permitted only because reanalysis turns the adjunct into a complement of the main verb. If so, c-command between pronoun and extraction position would be established, reducing the strong effect to Principle C: (73)

Williams (1994: 71–72) “Suppose there is a reanalysis rule that moves the adjunct from adjunct position to complement position. In that position it will be able to participate in feature passing under the definition of relativized head: (100) [[leave]VP [with t]]VP ⇒ [leave [with t]]VP “The reanalysis will not affect the fact that the adjunct is not an argument and so will not endow it with the ability to originate a scope index; but it will permit it to pass up scope indexes that originate in arguments within it. So the trace, which is an argument of the preposition with, will be able to pass up its scope index in the reanalyzed structure. The reanalysis rule must be regarded as a ‘marked’ possibility, to account for the semi-ungrammaticality of extraction from adjuncts.”

Williams’s description of this putative reanalysis is so terse and informal that I cannot see what the output structure would be in complex cases. As argued in Baltin and Postal (1996) and chapter 8, other common invocations of reanalysis fail remarkably, which suggests that great caution is in order about this one as well. Moreover, Williams’s specific reanalysis suggestion is untenable. First, if grammatical extractions from adjuncts depended on reanalysis of adjuncts as complements, such extractions would show the properties of extraction from complements. These include maximum freedom of category for extractees, whereas genuine extraction from adjuncts is extraction from selective islands (see Postal, 1998) and normally incompatible with most non-NPs, predicate nominals, and so forth. But extractions from the adjuncts at issue do not behave like extractions from complements but are only possible for a restricted range of nonsubject NPs. Compare previous object NP extraction examples with the bad cases of (74) that involve extractions of other elements: (74)

a. *the woman [for whom]1 the doctor told me that while buying the ring t1. b. *[How long]1 did the doctor tell you that while washing his hands t1?

JUNK SYNTAX

1

225

c. *[What kind of a specialist]1 did the doctor tell you that story long after becoming t1? d. *[Underneath no elm tree]1 did the doctor tell you that story after sitting t1.

These data show that taking the extractions from adjuncts to involve reanalysis as complements just gives the wrong answer. Second, if reanalysis existed in these cases, logic requires that it be either optional or obligatory. If obligatory, the good nonextraction case anaphoric linkages like those in (70a), (71a), and (72a) could not exist. Therefore, the putative reanalysis must be optional. But if it is in general optional, then the bad extraction anaphoric linkages would have nonreanalysis structures and so could not follow from Principle C on the structure where reanalysis is absent. The only way to avoid this conclusion would be to claim that the ‘backward pronominalization’ can only exist when there is no reanalysis but that extraction from the apparent adjuncts is only possible given reanalysis. This is in effect claimed by Williams, who generalizes to the assertion that backward pronominalization into an adjunct is uniformly banned in the presence of extraction from that adjunct. This entails that such binding is banned even when there is no potential strong effect. The only actual support cited for this claim is (75): (75)

Williams (1994: 72) ⊄What1 should I warn her2 before giving Mary2 t1?

But even if the anaphoric linkage in (75) is bad, which I doubt, I do not find that the relevant type of backward anaphora is in general incompatible with extraction. For me, the anaphoric linkages in all of (76) are fine: (76)

a. ⊂What1 they warned her2 sternly before providing Mary2 with t1 was a rocketpropelled bicycle. b. ⊂What1 did the doctor try to talk to her2 while poking Gladys2 with t1? c. ⊂[Which principle]1 did the professor make fun of them2 while lecturing [the firstyear students]2 about t1? d. ⊂[Which principle]1 did the professor make fun of them2 while lecturing about t1 to [the first-year students]2? e. ⊂It was orange sherbet that1 Sonia giggled at him2 while feeding t1 to [little Bobbie]2.

If correct, the data in (76) provide another reason that nothing like Williams’s proposal offers an alternative to the view that (70b), (71b), and (72b) are strong effects whose offending pronouns fail to c-command the extraction site.

4. The copy trace version 4.1. Remarks I turn to the copy trace version of the Principle C account of the strong effect. This conception of traces actually goes back at least to Chomsky (1981: 89–90), where it was considered to deal with the secondary strong effect problem but not adopted:

226


“The natural way to work this out in the present framework would be to establish the convention for Move-a that when a is moved it is not deleted but left unchanged, apart from a feature D indicating that it is to be deleted in the PF-component.”11 At first glance, this modification seems to eliminate an array of defects that haunted the empty category trace version. Specifically, it seems to give a reason that the relevant traces are R-expressions. Second, it seems to solve the problem of strong effects induced by PP extraction, seen in, for example, (26b), and with no special assumptions about PPs. Third, it seems to resolve the secondary strong effect cases, treating these essentially like Riemsdijk and Williams’s NP Structure account without the latter’s special assumptions. Fourth, the copy theory of A-bar traces seems to resolve the asymmetry issue. For, remarkably, it seems to incorporate without special cost a version of Barss’s (1986) modification of Chomsky’s original proposal, under which the trace of each moved category is of the same type with respect to the binding theory as the extractee. The copy trace variant of the Principle C proposal thus seemingly yields correct answers for the cases in (77), all of which except for a are misanalyzed by the empty category version:12 (77)

a. b. c. d.

⊄Who1 did you persuade him1 that Joan would marry t1 = [who1]? ⊂Him1, I persuaded Joe1 that Joan would marry t1 = [him1]. ⊂Herself1, I persuaded Mike that Joan1 would discuss t1 =[herself1]. ⊄[To Marsha1]2, I persuaded her1 to get Bill to talk t2 = [to Marsha1]2.

A violation ensues in (77a) but not in (77b, c), just as desired, since only in the former is the copy trace an R-expression. Moreover, a violation is properly specified for (77d), because the pronoun improperly binds the occurrence of Marsha inside the copy trace. Despite these improvements though, the proposal remains nonviable. Least seriously, it does not solve the defect of ‘crossed’ pronouns inside PPs, as in (26); the copy trace structure still fails to induce a strong effect violation under a c-command statement of Principle C for cases like (78): (78)

⊂ Marsha1, I mentioned to her1 that Bill was infatuated with t2 = [Marsha1]2.

And it also fails to induce a Principle C violation in CP topicalization cases like (15a). I consider other unresolved flaws presently. But a deeper issue is the independent tenability of the assumption that each extraction can be associated with a trace copy of the extractee. 4.2. The general untenability of the copy view of traces Real doubts about this claim should have arisen given that thirty-one years ago Perlmutter, in a rarely if ever credited work, arguably among the first associating tracelike elements with movements, suggested a view in which each extraction left a resumptive pronoun, under a different name; see (79):

JUNK SYNTAX

(79)

1

227

Perlmutter (1972: 73 ) “(1) Rules that ‘chop’ constituents over variables in the sense of Ross (1967) do not exist. (2) Rules that appear to be ‘chopping rules’ are actually ‘copying rules’ that leave behind a shadow pronoun in the position of the constituent that has apparently been ‘chopped.’”

Moreover, Perlmutter gave factual arguments for the pronominal character of the invisible elements. While works like Cinque (1975) and Hirschbühler (1975) soon probably refuted the general claim, nothing to my knowledge has ever refuted the particular assertion that French nonrestrictive relative extraction sites manifest pronominal properties. Despite this, that extraction type induces strong effects just as English nonrestrictive relative extraction does. So, in terms of the copy trace based Principle C theory, these French cases would have to be assigned R-expression traces, thereby failing to account for the pronominal characteristics Perlmutter documented. Somehow assigning them pronominal traces would account for Perlmutter’s observations but would fail to yield the strong effects. One need not depend on French; an argument with the same logic arises from observations about the class of English extractions referred to as B-extractions in Postal (1994a, 1998). These include NP topicalization, NP clefting, and nonrestrictive NP extraction. Using Perlmutter’s methodology, I have argued that these extractions manifest pronominal gaps. The factual basis for this claim is that the extractions in question, though not the other NP extractions called A-extractions, are incompatible with positions, called Antipronominal Contexts, that ban weak definite pronouns.13 At issue then are facts like (80)–(84), which reveal correlations between ill-formed B-extraction sites and antipronominal contexts. These sites are, however, critically, not incompatible with A-extractions: (80)

a. b. c. d. e. f.

Marshall painted his trailer green/*it. *Green1/[*That color]1, Marshall never painted his trailer t1. *It was that color that1 Marshal painted his trailer t1. *Green1, which1 Marshall painted his trailer t1, [Which color]1 did Marshall paint his trailer t1? the color that1 Marshall painted his trailer t1

(81)

a. b. c. d. e.

Jerome was speaking in Latin/*it. *[That language]1, no one was speaking in t1. *It was Latin that1 Herman was speaking in t1. *Latin1, which1 Herman was speaking in t1, . . . [Which language]1 was Herman speaking in t1?

(82)

a. b. c. d. e.

Joe met some other woman/*her yesterday who was telepathic. [Some other woman]1, Joe met t1 yesterday (*who was telepathic). It was some other woman whom1 Joe met t1 yesterday (*who was telepathic). Some other woman, whom1 Joe met t1 yesterday (*who1 was telepathic), screamed. [Which other woman]1 did Joe meet t1 yesterday (who was telepathic)?

228


(83)

a. b. c. d. e. f.

It was that death ray/*it that Mike used. *[Some other death ray]1 it was t1 that Mike used. *It was that death ray that1/which1 it was t1 that Mike used. *that death ray, which1 it was t1 that Mike used, . . . [Which death ray]1 was it t1 that Mike used? [Whichever death ray]1 it was t1 that Mike used, it didn’t work.

(84)

a. b. c. d. e. f.

The concert lasted the whole week/*it. *[The whole week]1, the concert lasted t1. *It was the whole week that1 the concert lasted t1. *The whole week, which1 the concert lasted t1, . . . [What length of time]1 did the concert last t1? [No matter how long]1 the concert lasted t1, . . .

These and numerous other cases of the same type argue that English B-extraction sites are linked to nonovert resumptive pronouns. But as is well known and has been touched on variously earlier, these B-extractions nonetheless induce strong effects when the extractee is of the type characterized as an R-expression in Principle C terms; see (85): (85)

a. ⊄[That woman]1, Joe persuaded her1 that you would hire t1. b. ⊄It was that woman whom1 Joe persuaded her1 that you would hire t1. c. ⊄That woman, whom1 Joe persuaded her1 that you would hire t1, is waiting.

So, as in the French cases Perlmutter studied, the copy view of traces is incompatible with evidence that shows the pronominal character of a certain proper subset of all extraction sites, including those of English B-extractions, a pronominal character not dependent on pronoun extraction. A parallel argument against the copy trace view is derivable from the observations in Cinque (1990) and Postal (1993a, 1994b, 2001a, 2001b) that involve parasitic gaps (P-gaps). These works argue that nominal P-gaps are pronominal in the sense just characterized.14 For instance, P-gaps are impossible in the antipronominal contexts of (80)–(84), as only partly illustrated in (86): (86)

a. b. c. d.

*[Whatever color]1 Marshall hated t1 after painting his trailer pg1, . . . *[What length of time]1 did Alice waste t1 while trying to prove the concert lasted pg1? [What woman]1 did Joe watch t1 play tennis who was telepathic? [What woman]1 did Joe discuss t1 while watching pg1 play tennis (*who was telepathic)?

Here the bad version of (86d) reinforces the point by showing the contrast between ordinary extraction, which is insensitive to antipronominal contexts, and P-gap extraction, whose gaps are pronominal. Despite that, P-gaps induce strong effects, as probably first noted in Barss (1986), and thus contrast with topicalized or clefted visible pronouns. In Principle C terms then, they should involve copy traces of the R-expression type. But this is incompatible with the pronominal nature of the gaps. These points are illustrated in (87):

JUNK SYNTAX

(87)

1

229

a. Barss (1986: 378) ⊄It’s John who1 Mary voted for t1 after he1 asked someone to nominate pg1. b. Cinque (1990: 150) ⊄Who1 did they find t1 hostile before he1 realized they wanted to help pg1? c. ⊄[What woman]1 did Joe discuss t1 while she1 tried to persuade Mike to hire pg1?

That (87) involves genuine strong effects is supported by the fact that the Asymmetry Property manifests, although there is one difficulty; see (88): (88)

Barss (1986: 377) ⊂It was himself that1 John1 nominated t1 before he1 voted for pg1.

The difficulty is that while Barss (1986: 378) claimed that the anaphoric linkage in (89a) that involved an extracted pronoun rather than reflexive was acceptable, for me it is impossible, as is that in (89b). But perhaps these judgments are linked to the fact that despite accepting things like (89a) and the short forms of (89b and c), I find the full versions of (89b and c) ungrammatical: (89)

a. Barss (1986: 378) ⊂Barss/⊄PostalIt was him that1 John1 claimed Mary liked t1 even though he1 knew she hated pg1. b. ⊄It was himself that1 John1 claimed Mary liked t1 (*even though he1 knew she hated pg1). c. It was himself that1 John1 claimed Mary liked t1 (*even though I knew she hated pg1).

When we put all these facts together and assume (89) is somehow resolved consistently with the Asymmetry Property, P-gaps are seen to be systematically pronominal and yet to induce strong effects, inconsistent with a Principle C account based on the copy trace view, which could only predict the strong effects by taking the relevant P-gaps to be R-expressions. An additional general problem for the copy trace view relates to the fact known since the 1960s illustrated in (90): (90)

a. ⊄She1 criticized some of the men who visited Betty1. b. Postal (1971: 85) ⊂[Which of the men who visited Betty1]2 do you think she1 criticized t2?

Left extractions of the sort that induce strong effect violations have the effect of eliminating certain pre-extraction anaphoric blockages that would be due to Principle C under Chomsky’s binding theory. Under a direct modification of (90b) in terms of the copy trace theory, the result, as in effect previously noted by Kuno, is (91), which wrongly induces a Principle C violation: (91)

See Kuno (1987: 63) ⊂[Which of the men who visited Betty1]2 do you think she1 criticized t2 = [Which of the men who visited Betty1]2

230


With no reference to Kuno’s criticism, it has been proposed by Chomsky (1995b) following Lebeaux (1991) that adjuncts, which include restrictive relative clauses, are introduced by generalized transformations in a way that avoids the violation, presumably by yielding a copy trace in (91) that does not contain Betty. Such proposals are supported by citing supposed contrasts from Riemsdijk and Williams (1981) and Freidin (1986) to the effect in (92): (92)

Chomsky (1995b: 204) a. ⊄[Which claim that John1 was asleep]2 was he1 willing to discuss t2? b. ⊂[Which claim that John1 made]2 was he1 willing to discuss t2?

I do not perceive this distinction and find the anaphoric linkages in both (92a and b) acceptable. And I find it even clearer that the anaphoric linkages are well formed in (93a and b), just as much as in the adjunct case (93c), a judgment supported by its consistency with (93d and e):15 (93)

a. ⊂[The claim that [the director]1 was corrupt]2, he1 was unwilling to discuss t2. b. ⊂[That [the director]1 was corrupt]2, everyone knew that he1 would always be able to deny t2 with a straight face. c. ⊂[The claim that [the director]1 made t2]2, he1 was unwilling to discuss t2. d. Ross (1973: 198) ⊂[That Ed1 was under surveillance]2 he1 never realized t2. e. Culicover (1997: 333) ⊂[That John1 had seen the movie]2, he1 never admitted t2.

Consequently, the complement versus adjunct distinction does not seem relevant, Lebeaux’s strategem does not work, the problem for the copy trace theory raised by the incorrect implications of (91) remains undealt with, and so yields a further objection to the copy trace theory. Still further direct problems for the copy trace view are found in (94), which involve reflexives that evidently some, including me but not all speakers, accept: (94)

a. Barss (1986: 276) ⊂It was himself whom1 John1 said Mary loves t1. b. *⊄Mike1 said that Gladys would never marry himself1. c. ⊂Mike1 said that himself1 Gladys would never marry t1. d. ⊂It was himself whom1 Mike said that Gladys would never marry t1. e. Lasnik and Saito (1992: 110) ⊂John1 thinks that himself1 Mary likes t1. f. Pollard and Sag (1992: 295) ⊂It was herself that1 Mary1 thought Bill admired t1 most. g. ⊂Herself1, I heard Barbara1 claim that Tony hated t1. h. ⊂Himself1, I couldn’t convince Harold1 to let Betty visit t1. i. ⊂It was himself that1 they told Jim1 to have Betty tutor t1.

For in these cases a copy trace view induces reflexive traces in positions where actual reflexives are impossible.

JUNK SYNTAX

1

231

Finally, recall the various different non-PP cases where the empty category view of traces permitted no reduction of strong effects to Principle C because the offending pronoun failed to c-command the extraction site. These were the coordinate, exceptive, noncrossover effect, and extraction from adjunct cases. Obviously, enrichment of the structure of traces has no effect whatever on this flaw; so the second version of the Principle C reduction idea inherits the defect of not inducing strong effects in a range of structures where they actually occur. I have tried to show then that despite a certain initial relative success compared to the empty category version, the copy trace version of a Principle C account ultimately cannot rescue the Principle C view of strong effects. More generally, I have attempted to indicate that the copy trace idea is itself untenable as a general conception of the nature of extractions, for various reasons, some touched on by Perlmutter more than thirty years ago.

5. Substantive conclusions To conclude, Chomsky’s claim (4) was clearly untenable when made, remained untenable through the period that preceded adoption of the copy view of traces, and is also untenable under that view, which is itself in general untenable. Moreover, given the problems the overall conception faces, which include a rich range of cases where strong effects exist in the absence of any c-command between pronoun and extraction site, the prospects for future successful revision seem extremely poor. Further, the Principle C account incorporates the idea that strong effects can, via appeal to traces, be reduced to the principle that bars anaphoric linkage between a form F and an anaphoric form A where A is found in a higher constituent than F. Considerable evidence was presented that this more general idea, which dates to the original trace proposals of Culicover and Wasow, is also wrong. If correct, given the limited current alternative descriptions of the strong effect, the conclusions show that the nature of the principles that truly underlie this phenomenon should be regarded as essentially open. Further, the Principle C account of the strong effect is often, as in (95), cited as evidence that supports the postulation of the nonpronominal traces on which it depends: (95)

Lasnik and Uriagereka (1988: 41) “Note that this account of so-called strong crossover demands that Condition C be applicable to derived structure, a conclusion we reached on other grounds in section 1.3.2. Furthermore, to the extent that this analysis is successful, it provides a second argument for the basic tenet of Trace Theory—that movement rules leave traces. If there were no traces, then there would be no R-expression for Condition C to constrain.”

Significantly then, our conclusions here support the claim that there is no real factual support for any tracelike objects connected with extraction except invisible resumptive pronouns of the sort originally posited by Perlmutter. For the latter I believe substantial evidence is available, in Obenauer (1984, 1985, 1986, 1992) as well as in Cinque (1990) and Postal (1994a, 1998), although this claim

232


is sharply challenged in Levine (2001). This support is of essentially the sort that, as argued in Sag and Fodor (1995), has never successfully been provided for nonpronominal traces.

6. Methodological conclusions A theoretical linguistic proposal like Chomsky’s Principle C hypothesis about the strong effect could have been, as was, I believe, the proposal in Postal (1971), wrong without being disgraceful But as the text and notes 4, 6, 9, 10, and 11 support, the moves to introduce claim (4), that Chomsky’s (1981) Principle C explained the strong effect, not only embody a factually and theoretically untenable grammatical proposal but also flout minimal standards of reasonable linguistic scholarship. In particular, it was shown that in addition to saturating the claim with theoretical and factual inadequacies that were not treated openly, no attempt was made even to keep claim (4) consistent with the original strong effect data in Postal (1971). The assertion ten years later that Principle C entailed the strong effect has been shown to have ignored (i) induction of strong effects by non-NP extractions, (ii) the secondary strong effect, and (iii) the Asymmetry Property, all of which were documented in Postal (1971).16 There is another aspect of the inadequacy of the development of (4) as a linguistic claim worth stressing. While it has been taken to be an actual linguistic discovery that the strong effect is an entailment of Principle C (given its supporting assumptions about traces, traces as R-expressions, etc.), it is notable that this claim exists only as scattered passing remarks in various works. Not only is there no complete section of any work, no article, no monograph devoted to its justification, there is hardly a full paragraph. Surely this fact, combined with the extraordinary shoddiness of the support for the proposal and the disdain for standards involved in its promulgation, should suggest a moral: When someone claims in passing to have a major theoretical syntactic result but can neither produce nor cite any work devoted to justifying the supposed accomplishment, one must rightly suspect that, as in this case, one is dealing with junk syntax.

8

Junk Syntax 2 “There Remain a Few As Yet Unexplained Exceptions”

1. Background During the now nearly five decades during which the ideas of generative transformational grammar have been applied to English syntax, passive constructions such as (1) have been a continuing focus: (1)

a. Claudia was interviewed by that company. b. Few linguists were considered by Marie capable of solving that problem.

If these ideas were sound, one could today reasonably expect a basic analysis that embodied a core of insight into English passive structures to have emerged. Such has often been implied, as in the implicit suggestion in (2), which yielded the title of this chapter, that the only problems that persisted (as of 1977; presumably even a few of those would have since been resolved) are at the margins: (2)

Chomsky (1977a: 14, note 14) “On the analysis of passive constructions, see Reflections on Language, Chapter 3. There remain a few as yet unexplained exceptions.”

I am aware of no repudiation since by the author of this remarkably optimistic view of the level of success achieved by his transformational movement view of English passivization.

233

234


Moreover, as is well known, in the overall Principles and Parameters tradition that represents the later evolution of the ideas of the author of (2), the description of passives continues to share a great deal with that of the period of the remark. It is as much standard doctrine now as in 1977 that English (verbal) passives are properly described transformationally, via (DP) movement of a specific type (now called A-movement); see, for example, Culicover (1997: 89–101) and, more recently still, Hornstein and Nunes (2002: 28–29). If so, the unrepudiated optimism of (2) should not be less tenable today than in 1977. But the burden of this chapter is twofold: first, to suggest that beyond (partially) capturing the gross semantic/selectional similarities between active object and passive subject and active subject and passive by phrase already noted in 1957 (see [3]) there is little about the relevant construction that transformational accounts get right; second, and more important, to argue that the exposition and justification of transformational views of passivization have been prototypical instances of junk syntax. It is argued that claim (2) was then and would be today not only wrong but also deeply irresponsible.

2. The beginning: syntactic structures The earliest generally available statement of a transformational view of English passives is represented by the informal account in (3): (3)

Chomsky (1957: 42–43) “Passive sentences are formed by selecting the element be+en in rule (28iii). But there are heavy restrictions on this element that make it unique among the elements of the auxiliary phrase. . . . “Finally, note that in elaborating (13) into a full-fledged grammar we will have to place many restrictions on the choice of V in terms of subject and object in order to permit such sentences as: ‘John admires sincerity,’ ‘sincerity frightens John,’ ‘John plays golf. . . . John drinks wine,’ while excluding the ‘inverse’ non-sentences’ ‘sincerity admires John,’ ‘John frightens sincerity,’ ‘golf plays John,’ ‘wine drinks John.’ But this whole network of restrictions fails completely when we choose be+en as part of the auxiliary verb. In fact, in this case the same selectional dependencies hold, but in the opposite order. That is, for every sentence NP1–V–NP2, we can have a corresponding sentence NP2–is + Ven–by + NP1. If we try to include passives directly in the grammar (13), we shall have to restate all of these restrictions in the opposite order for the case in which be + en is chosen as part of the auxiliary verb. This inelegant duplication, as well as the special restrictions involving the element be + en, can be avoided only if we deliberately exclude passives from the grammar of phrase structure, and reintroduce them by a rule such as: (34) If S1 is a grammatical sentence of the form NP1 – Aux –Y– NP2, then the corresponding string of the form NP2 – Aux + be + en –V– by + NP1 is also a grammatical sentence.

JUNK SYNTAX

2

235

“For example, if John–C–admire–sincerity is a sentence then sincerity – C + be + en – admire – by + John (which by [29] and [19] becomes ‘sincerity is admired by John’) is also a sentence.”

This passage introduced one clearly correct insight of the transformational description—namely, the existence of corresponding selections. Beyond that, from the beginning there was vast exaggeration, a remarkable detachment from the facts of English, as well as outright distortion—that is, aspects of junk syntax. Most significant in particular is the claim that for every sentence of the form [NP1 V NP2] there is a corresponding passive of the form [NP2 is + Ven by NP1]]. One might assume today in defense that this long-ago falsehood (see later discussion) was due to mere naïveté, carelessness, or the inevitable limits of a beginning field. But consider passage (4) from the author’s thesis of 1955, a study finished two years before the publication of the volume from which (3) is drawn:1 (4)

Chomsky (1975a: 565) “There are many exceptions to the transformations that we have set up . . . “We might mention several incidental exceptions (whether real or apparent, only future investigation can determine) to the transformations we have constructed. As instances of actives with no corresponding passive we have: 471

(a) (b) (c) (d) (e) (f)

this costs a lot of money this weighs three pounds John traveled three days Mary married John misery loves company he got his punishment

(g) (h) (i) (j) (k)

he had an accident no one foresaw any improvement he didn’t like either of them he only likes certain people the artist redecorated it completely”

That is, the author was aware several years before the publication of claim (3) in 1957 that it was false, had even said that there were many exceptions to his transformations in general, and knew in particular that his passive mapping failed for, for example: (5)

a. [NP John] [traveled] [NP three days] ⇒ b. [NP this] [costs] [NP a lot of money] ⇒ c. [NP he] [had] [NP an accident] ⇒

* three days were traveled by John * a lot of money is cost by this *an accident was had by him

Further, Chomsky noticed (1975a: 534) that (6b and d) were ungrammatical, which means that the claim in (3) was also false for (6a and c), the latter example explicitly mentioned: (6)

a. b. c. d.

John saw himself in the mirror. *Himself was seen by John in the mirror John ate dinner by himself. *Dinner was eaten by himself.

More generally, as amply (but still only in part) illustrated later, the degree to which the ‘every’ claim fails is extraordinary. One can only speculate as to why such

236


a deliberate falsification was advanced in (3). But one can hardly avoid consideration of the fact that it appears in the author’s first published attempt to tout the virtues of transformational grammar. Nor can one ignore the truism that a view always looks better if it can claim exceptionless principles, rather than rough approximations. Putting these together seems not to leave much mystery as to the ground for the falsification. In any event, the junk linguistic feature of outright fact distortion to render a proposal more attractive than it is had evidently surfaced in the first work that advocated a transformational view of English passives. Moreover, claim (3) also contains the unsupportable assertion that the duplication of selectional and other restrictions that Chomsky had argued were associated with a phrase structural account of passive cases could only be avoided via a rule like the author’s (34). This linked to similar exaggeration on page 78, where it was claimed that “any grammar that can distinguish singular from plural is sufficiently powerful to enable us to prove that the passive requires inversions of nouns phrases,” and on page 79, where it was stated that “[s]uch verbs prove quite conclusively that the passive must be based on an inversion of subject and object.” Critical is the occurrence of the strong ‘prove’ and still stronger ‘prove quite conclusively’ in these claims. The unfounded, exaggerated character of such remarks is shown entirely internal to the development of the author’s own views. For he has, of course, long since abandoned the idea that there is inversion of DPs in passives and only recognizes movement of objects; see section 5. So what was “proved (quite conclusively)” by X in 1957 was not even considered true by X a couple of decades later. Evidently there was no proof but only empty claims of such, raising minimally another issue of conduct. For the later abandonment of the conclusion was not to my knowledge associated with any public retractions of the false claims. A distinct strand of junk linguistic practice is illustrated. Claim something quite strong, using words like prove, later abandon the claim if it proves a burden, but never discuss or admit the lack of basis for the earlier admittedly false and exaggerated assertion. And there was more: (7)

Chomsky (1957: 82–83) “As another example of a similar type, consider the sentence (108) John came home. “Although ‘John’ and ‘home’ are NP’s, and ‘came’ is a Verb, investigation of the effect of transformations on (108) shows that it cannot be analyzed as a case of NP – Verb – NP. We cannot have ‘home was come by John’ under the passive transformation, or ‘what did John come’ under the question transformation T. We must therefore analyze (108) in some other way (if we are not to complicate unduly the description of these transformations), perhaps as NP – Verb – Adverb. Apart from such considerations as these, there do not appear to be very strong reasons for denying to (108) the completely counterintuitive analysis NP – Verb – NP, with ‘home’ the object of ‘came.’”

In the discussion, the author encounters a failure of passivization, represented by (8b):

JUNK SYNTAX

(8)

2

237

a. John came home. b. *Home was come by John.

Passage (7) does not indicate that the problem is general across the class of intransitive verbs that accept a postverbal home: (9)

a. John crawled/dashed/flew/hopped/ran/swam home. b. *Home was crawled/dashed/flown/hopped/ran/swam by John.

The author’s reaction to (8b) was remarkable. Although the passage already says that John is an NP, came is a verb, and home is an NP, it then claims bizarrely that (8a) cannot be analyzed as a case of [NP – Verb – NP], because if it is, it is wrongly predicted that, for example, (8b) is good. It is suggested that it should be given some other analysis, perhaps [NP – Verb – Adverb], to block the bad case. This proposal is incoherent. Stating that home was an NP had already granted that (8a) has the analysis [NP Verb NP].2 Thus one need not depend on passages from Chomsky (1975a). Under its own constituency assumptions, the 1957 volume itself contains a counterexample to its claim in (3) that every [NP1 V NP2] structure yields a passive. It is, moreover, irrelevant to any current point whether it is claimed that in addition the structure of cases like (8a) includes a subtree of the form (10), fig. 8.1: (10)

Adverb

NP.

 8.1.

For the author’s conceptions of transformational grammar have always been such that such a subtree would not block application of a rule defined in terms of NP – Verb NP.3 The only conclusion then is that the author had, as in the case of the counterexamples in (4), already noticed another anomaly for his transformational view of passive in cases like (8b) and had no coherent solution for it but failed to make that situation explicit. As an account of the relation between English active and passive structures, (3) was clearly radically false; see sections 4–6 for an extensive listing of counterexamples to it. Moreover, other classes of data, while not counterexamples, also show its extreme inadequacy. These involve uncontroversial passives that do not correspond to actives of the form cited in (3). The most visible group of such seemingly exceptional cases are the prepositional passives or pseudopassives. Section 6 treats these exceptions to principles like those embodied in (3).

238


Evidently then, fundamental junk linguistic features were already highly visible in the work that introduced a transformational description of English passives to the general linguistic public. Despite this, the view that there is a viable, moreover insightful, transformational account of English passives had great fairly immediate popular success, which has persisted to the present day. The implication is that just because a junk linguistic proposal is awful linguistics or offends decent standards of inquiry, that doesn’t begin to determine that it will not be widely accepted. Parallels are clear with the remarks in chapter 7 about the social success of the Principle C account of strong crossover phenomena despite its incompatibility with data published a decade before that account surfaced.

3. “Applies blindly” The falsehood in (3) that every English [NP + V + NP] structure yields a corresponding passive received, in effect, a more elaborate and abstract formulation by its author during the 1970s. Associated with that development was an unheralded, repeated appeal to protective devices, elements that in effect can defend the formulation against any data. This is illustrated in (11): (11)

Chomsky (1973); reprinted in Chomsky (1977a: 82–83) (emphasis mine: PMP) “[T]he Passive transformation (reducing it to essentials) applies to any phrase marker that can be ‘factored’ into five successive substrings in such a way that the second and fourth are noun phrases, the third a verb of a particular category (perhaps determined by some semantic property), and the first and fifth anything at all (including nothing). Thus the structural condition defining the transformation can be given in the form (Z, NP, Vx, NP, Y). The transformation rearranges the noun phrases in a fixed way. It will, therefore, apply to the phrase markers underlying the sentences of (1), converting them to the corresponding passive forms: (1)

a. b. c. d. e.

Perhaps-John-read-the book-intelligently John-received-the book John-regards-Bill-as a friend John-painted-the wall-gray John-expects-the food-to be good to eat

“Evidently, the semantic and grammatical relation of the main verb to the following noun phrase varies in these examples (there is no relation at all in [e]), but these relations are of no concern to the transformation, which applies blindly in all cases.”

Although position (11) offers a refined formulation of the ideas of the Syntactic Structures account, it is similar in most key ways. Passivization continues to be characterized via NP/DP movement of both underlying postverbal DP and subject DP. But there are new emphases; specifically, transformations are stipulated to be blind to semantic and grammatical relations. So passivization in particular is thereby taken to be independent of these features.4 The highlighted claim of blind application only appears strong since the associated hedge that a verb-instantiating term three of the

JUNK SYNTAX

2

239

structural description has to be “of a particular category,” not characterized (moreover, one possibly determined by some unspecified semantic relation), renders claims of blind application vacuous. Any potential counterexamples, for example, those of (5) or (8b), could be handled merely by stipulating that their verbs are not of the right category. Such assignments could never be wrong since no independent criteria for membership in that category were presented. So, as a conception of the conditions that permit an English active structure to have a corresponding passive, proposal (11) embodies only the tautology that for a fixed active A, A has a corresponding passive or not, depending on whether A’s verb belongs to category Vx or not.5 As (12) documents, the author of (11) made similar or related pseudostrong claims on other occasions during the 1970s, which indicates that the position cannot be dismissed as a temporary aberration: (12)

(all emphases mine: PMP) a. Chomsky (1971: 30–31) “In all cases, the passive is formed by the rule informally described a moment ago. The rule pays no attention to the grammatical and semantic relations of the main verb to the noun phrase that follows it. Thus in ‘I believed your testimony,’ the noun phrase is the grammatical object of ‘believe.’ In ‘I believed your testimony to be false,’ it bears no relation to ‘believe,’ and is the subject of ‘be false.’ In ‘I believed your testimony to have been given under duress,’ it bears no relation to ‘believe’ and is the grammatical object of the embedded verb ‘give.’ Yet in all cases, the rule applies blindly, caring nothing for these differences.’ Thus in an important sense, the rules are structure-dependent and only structuredependent. Technically, they are rules that apply to abstract labeled bracketing of sentences (abstract, in that it is not physically indicated), not to systems of grammatical or semantic relations.” b. Chomsky (1972b: 118) “The basic property of transformations is that they map phrase-markers into phrasemarkers. Each transformation applies to a phrase-marker on the basis of the formal configurations expressed in it, and quite independently of the meanings or grammatical relations expressed by these formal configurations. Thus such sentences as John received the book, John read the book, John expected the book to be good, John gave Bill the book, and so on, undergo the passive transformation in exactly the same way. The transformation applies blindly to any phrase-marker of the proper form, caring nothing about meanings or grammatical relations. This situation is typical; I know of no exceptions, and no counterarguments that amount to more than terminological revision.”

While no insight into passives is visible in (11) or (12), they do reveal something about general principles of junk linguistics. These involve the tactic of making proposals seem significant by stating them in a vividly strong form (‘applies blindly’/ ‘independent of grammatical relations’) while at the same time covering oneself against potentially embarrassing counterexamples via background tempering with protective hedges that render them safe from any data. The particular hedges in (11) are among a variety of mechanisms invoked by its author to eliminate testability of

240


an otherwise strong claim about a transformational description of English passives. Other such devices with similar functions will be touched on presently. Hedging the sort of transformational views promoted in the 1970s so that they could not be falsified was well motivated. Despite boasts like (2), considerable data incompatible with an unhedged formulation had accumulated by the time claims like (11) and (12) were made. For example, Fillmore (1965), a well-known and prominently reviewed work of the time (see Kuroda, 1968; and Wilson, 1966), had discovered several generalizations about one type of English, his own (and, incidentally, mine). The generalizations concerned double object or ditransitive cases like (13): (13)

a. Armand sold her the six Uzis. b. Armand bought her the six Uzis.

Fillmore observed that in the type of English at issue, structure (13a), call it a type A ditransitive, corresponds to two distinct passives, while structure (13b), call it a type B ditransitive, corresponds to none. (14)

a. She was sold the six Uzis by Armand. b. The six Uzis were sold her by Armand.

(15)

a. *She was bought the six Uzis by Armand. b. *The six Uzis were bought her by Armand.

Fillmore noted, moreover, that it was in significant part predictable whether a double object structure was of type A or B. The type A variety was such that the first object alternated with a structure with a PP in to, while the type B was such that the first object alternated with a structure with a PP in for. (16) a. Armand sold the six Uzis to her (for Rhonda). b. *Armand bought the six Uzis to her (for Rhonda).

Evidently, nothing in proposals (11)/(12) offers any insight into these facts. Moreover, (15a, b), and so on, are further prima facie counterexamples to the claim in (3) about “every” transitive active and to any factual content to a claim that there is some rule that “applies blindly.” And (14b), an instance of the tertiary passives of Postal (1986a), challenges the assumption that only directly postverbal DPs passivize in English. If those assertions had been serious, they would have had to take into account Fillmore’s observations and advance some proposal to keep the double object data consistent with the formulation. But typically for junk linguistics, formulations like (11)/(12) were advanced, multiply published, and defended as if Fillmore’s observations, uncited, had never been made. And it is entirely impossible that the author of (3), (11), and (12) was unaware of the dual passive possibilities of cases like (14). Least of all, it was traditionally documented; see, for example, Jespersen (1927: chapter 15). Moreover, amazingly, he had himself noted this fact before Fillmore. The point is made explicitly in his thesis, noted earlier to have preceded Fillmore’s work by a decade, despite its pub-

JUNK SYNTAX

2

241

lication delay by the same amount. Chomsky (1975a: 493) cites the type A ditransitive cases: (17)

a. he was given several books by the teacher. b. several books were given him by the teacher.

Worse: (18)

Chomsky (1975b: 242, note 43) “In ordinary passives such as (55), the rule of NP-preposing disregards the grammatical relation between the verb and the NP following it, at least if we use the term ‘grammatical relation’ in something like its traditional sense. Thus in (55), the rule moves the direct object, but in such cases as ‘John was told to leave’ or ‘John was promised that he would get the job,’ it is the indirect object that is preposed (cf. ‘John was told a story,’ ‘a story was told to John’; ‘our promise to John . . .’. . .).”

That is, a decade after the appearance of Fillmore (1965), the author of (11) and (12) cited double object passives selectively as supporting his view that passive movement ignores grammatical relations, with no mention of his own or Fillmore’s earlier observations, work that minimally is problematic for that view. This was feasible only by carefully picking examples of type A ditransitives and none of type B manifesting, as Fillmore documented, unpassivizable direct and indirect objects, and arguing that passivization does not disregard the type A/type B distinction, which nothing in the structures the author was offering distinguished. Moreover, it was necessary as well to avoid mention of type A tertiary passive examples like (14b) and (17b), since their citation would have inevitably raised issues for which no solutions were available. So where genuine linguistics would have manifested a real attempt to handle the facts Fillmore had discussed, some of which the author of (18) had noted even earlier, one finds instead spurious boasts like (2). The fact that claims like (11), (12), and (18) ignore the relevant ditransitive observations is in itself an excellent indication of the junk linguistic character of the passive proposals at issue. More generally, the history of generative discussion of passives like (14b) is arguably an exceptional illustration of the grip of junk linguistics. Although such examples are commonly cited in traditional descriptions of British English (see Curme, 1931: 117; Jespersen, 1927: 301–312; and Sweet, 1891:113) and were analyzed for American English in Fillmore (1965), Oehrle (1976), and Postal (1986a), many linguistic works of the last twenty years, especially more recent ones, deny the existence of American English passives like (14b); see, for example, Boeckx (1998, 451– 452), Bresnan (1982c: 25–29; 2001: 316), Lasnik (1999a: 198), Ouhalla (1994: 175), Pesetsky (1995: 124), Riemsdijk and Williams (1986: 117), Runner (1999: 155), and Ura (2000: 244–248). Evidently, many Americans, especially younger ones, do not accept them. But for many American speakers, including me, many examples are perfect. Example (17b) shows that the inventor of transformational ideas about passives being criticized here was in 1955 among them. And Christopher Potts kindly provides the following published written examples:

242 (19)


a. He was a man of many holdings—many of them handed him, as each ancestor finished his life’s run and passed the stick forward, handed him in a lineal descent of bonds and a bank, of glass birds and dishes, land, houses, and attitudes. M. Malone. Dingley Falls (p. 88) b. The young men crunched ice cubes and wolfed cheese sandwiches brought them by Chris Henry. M. Malone. Dingley Falls (p. 127) c. Lagniappe: a small gift given a customer by a merchant at the time of purchase. Webster’s 3rd New International Dictionary d. Included meanwhile in the epic consolations given the conservationists . . . were more than two million acres along the Yukon between Eagle and Circle. John McPhee. Coming into the Country (p. 246, hardcover) e. The friendship extended him by his classmates. Richard Russo. Mohawk (chap 16, p. 1) f. Cooper (1983) is the most explicit and fully developed attempt to pursue the first strategy, but the semantic techniques involved are significantly more complex than those afforded us by PC. Gennaro Chierchia and Sally McConnell-Ginet. Meaning & Grammar (p. 119)

So, given that many speakers reject them, a serious treatment of tertiary passives should have, even ignoring British English, throughout dealt with what are clearly partially contrasting grammatical systems. A proper account of American English passives evidently needs to specify how these differ, a point explicitly made in Bach (1980: 325) and Iwakura (1987: 94). Instead, it has, probably under the pressure of various theoretical assumptions, come to be widely accepted in generative circles, despite clear evidence to the contrary, that tertiary passives do not exist in American English. To see the truly bizarre character of this, one finds that, for example, Stroik (1997: 43), a work that explicitly develops the recent (‘minimalist’) transformational ideas of Chomsky (1993, 1995a, 1995b), does not flinch from deriving support for specific assumptions from their ability to predict the impossibility of such passives. That is, in 1997 one finds someone putatively developing the theoretical ideas of author A and claiming support for a version of those ideas from the purported fact that they predict the impossibility of sentences richly exemplified in the literature and that A himself had documented for his own NL in 1955. Perhaps even more tellingly, in a work submitted to a Web site of papers to honor the author of (18), Romero (1998), someone associated with that author’s own institution, asserts as part of a theoretical discussion of passives crosslinguistically that only ‘goal’ passives are possible in English ditransitives and cites: (20)

a. Almodovar was given the awards b. *The awards were given Almodovar

So here someone is “honored” by theoretical discussions that assume the essential contrary of what the honoree had documented (in [17b]) forty-three years earlier. This might be said to go beyond junk linguistics to a form of unintended parody.

JUNK SYNTAX

2

243

One further point about the interaction of ditransitive structures and passivization is worth highlighting. A fact rarely discussed in transformational terms is the existence of passive constituents embedded under get. Notably, while, for example, (18) correctly cites various ditransitive first (indirect) objects as being passivizable and (17b) correctly indicates certain ditransitive second objects are, even those that have that property in standard settings with be lack it in get cases: (21)

a. b. c. d. e. f. g.

The wrong books were/got ordered. The house was/got destroyed. The books were/got sold to Mike. Mike was/*got sold the books. Mike was/*got told several stories. Several books were/*got given him by the teacher. Nobody wants to be/*get sent threatening letters.

A parallel point holds for passive constituents embedded under causative verbs like have: (22)

a. Stella had the wrong books ordered. b. Stella had the house destroyed. c. Stella had the books sold to Mike. d. *Stella had Mike sold the books. e. *Stella had Mike told several stories. f. *Stella had several books given him.

Needless to say, nothing in the proposals about transformational descriptions of English passives accounts for these gaps. No aspect of (3), (11), or (12) seems to have any application. Indicated, I believe, is that the standard idea that all English passivization is to be reduced to a single undifferentiated operation is misguided. But in any event, the contrast between unsupported dreams like (2) and the factual reality of English is again quite palpable. Statements (11) and (12) assert that transformations, specifically that involved in passivization, apply blindly and are indifferent to grammatical relations. But consider: (23)

Chomsky (1971: 29–30) “Consider next the sentence ‘I believe the dog’s owner to be hungry.’ Applying the postulated operation, we locate the main verb ‘believe’ and the noun phrase ‘the dog’ following it, as before, and form ‘The dog is believed’s owner to be hungry.’ Obviously, this is incorrect. What we must do is choose not the noun phrase ‘the dog,’ but rather the noun phrase of which it is a part, ‘the dog’s owner,’ giving then: ‘The dog’s owner is believed to be hungry.’ The instruction for forming passives was ambiguous: the ambiguity is resolved by the overriding principle that we must apply the operation to the largest noun phrase that immediately follows the verb. This, again, is a rather general property of the formal operations of syntax.”

Here a really blindly applying rule would wrongly allow the impossible passivization of the genitive phrase. But instead of admitting that talk of blind application

244


was vacuous, the author invokes an auxiliary hypothesis, in effect, the A-over-A principle. But if one is allowed to invoke some new curative hypothesis every time blind application fails, the claim obviously has no testable consequences. Moreover, it was untenable even in 1971 to invoke a general A-over-A principle. Ross (1967/ 1986) devoted a whole chapter to the general untenability of such, work that Chomsky is silent about. But the really key point is that the failed passive of the genitive discussed in (23) is exactly the sort of case that is properly excluded if passivization is, contra the claims, sensitive to grammatical relations—in particular, if one views passivization in anything like traditional relational terms such as those in (52) ahead. For in no one’s sense is the genitive phrase any kind of object of the verb in whose clause it is unpassivizable. So, like the “applies blindly” claim, the supposed independence of passivization from grammatical relations was also interpreted in such a way as to admit of no falsification, at the same time it was proclaimed with great emphasis. At a certain point, it was assumed that the transformational DP preposing putatively crucial for characterizing verbal passives was the same operation as one that putatively yielded certain prenominal genitive DPs; see Fiengo (1977, 1980) and Hoekstra (1984: 133–136). So, for example, (24a and b) were both taken to involve preposing of a DP, a theory-internal way of saying both are passive structures:6 (24)

a. The city was destroyed by the giant gorilla. b. the city’s destruction by the giant gorilla

But Chomsky (1975b: 242, note 41), citing Joseph Emonds, noted a contrast between seeming nominal and verbal passives, as follows: (25)

a. the lecture yesterday ⇒ yesterday’s lecture b. he lectured yesterday ⇒ *Yesterday was lectured by him.

He then added: (26)

Chomsky (1975b: 242, note 41) “I think that in many cases, perhaps all, the discrepancies can be attributed to other factors.”

Crucially, the author does not say that his formulations of the NP preposing rule are wrong. Again, for a rule said to “apply blindly” the fact that it fails to apply in a case where its formulation says it should apply is not taken to show that the “apply blindly” claim is wrong. Rather than abandon the claim, in the particular case of (25), the author appeals instead to something he calls the subject-predicate relation, hypothesizing it to be defined on the surface structures of sentences but not on nominals: (27)

Chomsky (1975b: 242, note 41) “[I]t might be plausibly argued that ‘was lectured (by NP)’ is not a possible predicate of ‘yesterday’ accounting for the ungrammaticalness.”

JUNK SYNTAX

2

245

In effect, (27) claims that some semantic principle that involves predication blocks (25b) but is inapplicable to the good (25a) because the subject-predicate relation is undefined for NPs/DPs. Nothing about this proposal is serious. First, the author has not defined the putative subject-predicate relation or given any evidence that there is such a thing and not shown independently a single fact it accounts for. So there is no visible justification for assuming any such subject-predicate relation defined on surface structures. Rather, the contrary is true, given, for example, (28a, b), where grammatical passive subjects are expletives, meaningless phrases of which it would make no sense to predicate anything: (28)

a. It was proved that Disneyland is toxic. b. There are assumed to be space aliens in Congress.

Still worse for the author’s proposal, there are perfectly fine English sentences that have exactly the sort of subject NPs that are bad in passives like (25b): (29)

a. Yesterday found Mike in Detroit. b. Yesterday saw the Yankees lose to the Red Sox. c. Yesterday was found to be a poor day to launch the Space Shuttle.

What reason could there be that it is correct to predicate found Mike in Detroit, saw the Yankees lose to the Red Sox, or was found to be a poor day to launch the Space Shuttle of yesterday but not was lectured by him? None is of course given. In short, the casual, unsupported attempt to defend the NP preposing rule in Chomsky (1975b: 142) was entirely spurious and no genuine basis whatever was offered for the ungrammaticality of (25b). Nonetheless, the same account is in effect repeated in Chomsky (1977a: 177–178). The very appeal to it shows clearly the emptiness of claims about movement rules “applying blindly.” If the term ‘junk linguistics’ did not exist, it would arguably need to be invented for this invocation of an unmotivated and indefensible semantic principle of predication just to block the application of a rule that, if the “‘applies blindly” terminology had any content, would have to apply even if the predication principle were real.7 Moreover, the procedure of proposing a desperate, ad hoc patch for an otherwise factually leaking proposal about passivization seen in (27) is not unique. That junk linguistic characteristic recurs; when the author notes factual problems for his transformational approach to passivization, he appears never to contemplate even the possibility that these show it is false. A second highly revealing instance is found in Chomsky (1981: 147, note 108), which considers (30): (30)

They forced John to wait.

This is taken to be a control structure, in the author’s terms, one with an invisible PRO complement subject coindexed with one in the including phrase, here John. One assumes this would be taken to be the right analysis for the whole range of force + infinitive cases like, for example, (31):

246 (31)


a. Mike forced Selma to do the dishes. b. Germany forced Holland to surrender.

Suspiciously though, the author gives no argument for a control analysis and considers no alternatives. Moreover, the various novel tests appealed to in chapter 2 to differentiate control from raising structures unambiguously indicate that infinitival complements with force are raising, not control, cases: (32)

a. Middles The occupying power forced Uruguayan police to bribe cheaply. b. Metonymous Stock Structures Not even Zeus could force Lucent Technologies to go up. c. Partial Control *The director forced his subordinate to meet at noon.

That is, it was seen that control complements could not be middles or the metonymous structures and that raising complements could not manifest partial control. In all three respects then, the force complements behave like raising cases. The arguably wrong control assumption putatively leads to an expectation, not explained, that cases like (33) are bad: (33)

a. they forced it to rain (by seeding the clouds). b. the forced better care to be taken of the orphans (by passing new laws).

But according to the author (I agree) these seem “moderately acceptable.” A straightforward reaction would have been to simply reject the control analysis, the source of the false expectation about (33). But this idea was not even considered. Instead, it was proposed that these examples “are only derivatively generated (in the sense of Chomsky (1965, p. 227; 1972 pp27f.).” The former reference yields a discussion that claims that an adequate grammar directly generates all the perfect sentences but “derivatively generates” imperfect ones with, in some unspecified way, an account of how they are imperfect. The latter offers parallel claims with indication that what is being talked about is to be interpreted in part in terms of analogy, in terms of speakers “failing to take note of a certain distinction of grammaticalness.” And the author goes so far as to insist: (34)

Chomsky (1972c: 28) “There is no doubt that such processes of derivative generation exist as a part of grammar in the most general sense.”

One observes that (33) represents further cases where the sort of “blind application” views of passivization advocated by the author, if interpreted substantively, fail, since the following are bad: (35)

a. *it was forced to rain. b. *better care was forced to be taken of the orphans.

JUNK SYNTAX

2

247

But, as in his discussion of the contrast between DP preposing in clauses and nominals, instead of seeing some problem with his formulation, the author again seeks an ad hoc solution, this time, in effect choosing to criticize the examples and to exclude them from the NL or to claim that they involve violations that speakers are failing to notice. Despite the undefended and indefensible (34), there is evidently no reason to see this as anything but junk linguistics. Support for that conclusion is that the author gives no substance to the claim that (33) is ‘derivatively generated’. How is this done? What principle or aspect of the grammar do they putatively violate? No answers are offered. But the author does go on to offer what is called “evidence supporting this conclusion,” namely, the fact that they resist “further grammatical operations,” this a reference to the ungrammaticality of passives like (35). This putative evidence would have to have a logic of the following form: (36)

a. There are structures, for example, (33), putatively ‘derivatively derived’, that satisfy the input conditions for the author’s passivization (DP-preposing) rule. b. The output of that rule on the cited structures is ungrammatical. c. Fact b. supports the ‘derivatively derived’ claim for the structures of (33).

But no logical connection between premisses and conclusion is visible here; one could hardly better instantiate the notion ‘non sequitur’. Further, what is stated in (36) is that an operation, once said to apply blindly, gives as output from a grammatical input structure that satisies its requirements something ungrammatical. Instead of concluding from that that the rule is wrong, the author leaps to a view of the input in entirely obscure and unanalyzed terms as ‘derivatively derived’. But evidently state of affairs (36b) is exactly that which could show that a factual claim involved in the transformational passivization rule is false. Again then the author makes ominously clear that in his methodology nothing will be allowed to lead to that conclusion for a proposal he favors. Instead, he decided to attempt to marginalize the facts under the obscure rubric ‘derivatively derived’, making it at least rhetorically unnecessary to deal with them seriously. The thoroughly junk linguistic character of that move is shown differently by the fact that its logic, (36), would require taking every passive failure to indicate ‘derivative generation’ of the input. For some idea of the scope of the absurdity of this, see in particular section 7. Two additional cases would be: (37)

a. b. c. d. e. f.

The chimp has a peach. ⇒ *A peach is had by the chimp. Therefore, a. is ‘derivatively derived’. Wanda got a puppy. ⇒ *A puppy was got(ten) by Wanda. Therefore, d. is ‘derivatively derived’.

Moreover, as remarked in Postal and Pullum (1988: 657), Chomsky has himself not in general followed that logic and has on multiple occasions there referenced concluded merely that a passive failure indicates the existence of some linguistic

248


constraint that blocks the output. Obviously, that is what he did with respect to the nominal/clausal contrast of (25). The junk linguistic character of methodology (36) then includes the fact that its advocate refuses to even follow it consistently. The author then moves on from (33) and (35) to examples like (38): (38)

they prevented it from raining.

Here he says bizarrely (1981: 147, note 108) that it “should also be ungrammatical if prevent assigns a q-role to its object.” This is bizarre because even if one adopts a framework that sanctions concepts like ‘assign a q-role’, no independent reason is cited to make the assumption for the particular case and there is none: (39)

a. b. c. d.

He prevented tabs from being kept on her movements. God prevented it from being nice in the Congo. Only the president can prevent there from being a strike. The director prevented it from being revealed that the treasury was empty.

That is, the post-prevent position allows all sorts of DPs (highlighted) that the author has otherwise assumed cannot receive q-roles. Moreover, the novel tests that distinguish raising from control complements again indicate that the prevent cases are raising structures: (40)

a. Middles The occupying power prevented Uruguayan police from bribing cheaply. b. Metonymous Stock Structures Not even Zeus could prevent Lucent Technologies from going down. c. Partial Control *The director prevented his subordinate from meeting at noon.

So the author’s unsupported factual assumptions were clearly wrong. It was then claimed there is a parallel between his (35) and the force cases because supposedly the corresponding passive is again impossible: (41)

*it was prevented from raining

Here, though, as discussed in Postal and Pullum (1988: 657), the facts are wrong, since in general speakers find passives like (41) and (42) essentially perfect: (42)

a. Only through the use of nuclear weapons can it be prevented from raining. b. It could not for long be prevented from being noticed that he was dead.

The author says about (38) that “it has been proposed as an argument for raisingto-object,” referring without citation to one of the arguments in Postal (1974). But Chomsky asserts that no argument can be based on it “since the rules for generating it would appear to be idiosyncratic.” This illustrates a further junk linguistic lack of logic for several reasons. First, since no actual rules are given or cited, there is no

JUNK SYNTAX

2

249

way to evaluate any claim about them. If there is no such set or if its putative members were not correct, nothing could follow. But second, even if the author had in mind a relevant set of correct and idiosyncratic rules, nothing would follow. For no known logic determines that arguments can’t be based on idiosyncratic rules. They can be based on any facts and, even, in the case of reductio arguments, on false assertions. The “no argument” claim is entirely junk linguistic bluff. Thus the passages just discussed show again the emptiness of claims of blind application But they also illustrate the more harmful characteristic that when faced with data incompatible with some putatively significant claim, the author systematically refuses to contemplate its falsehood and instead has sought to invoke the most ill-defined and far-fetched protective moves, instantiating a solid strand of junk linguistics. A variety of other proposals that either inoculate his passive proposals from falsification or show that claims of “blind application” are empty have been made by the author of (11) and (12). Besides those already touched on, others are found in the list in Postal and Pullum (1988: 656–657).8 Overall then, the transformational passive proposals of the 1970s I have been discussing had the following junk linguistic characteristic: strong, prestigious-sounding claims systematically bound to one or another hedging device that guaranteed their factual emptiness. Not surprisingly, such spurious proposals coexisted with a flagrant disregard of well-documented and publicly known facts, such as those from Fillmore (1965), some even earlier noted in Chomsky (1975a).

4. Evolution of ideas The accounts of transformational description discussed so far, (3), (11), (12), and so on, all fall within an earlier set of transformational assumptions. From the mid-1970s through the mid-1980s, the theoretical framework that underlay these descriptions underwent significant evolution. At least six innovations are potentially relevant to subsequent discussion of passives in transformational terms: (43)

a. Each transformational movement was ultimately taken to leave an invisible trace in the position of origin. b. It was ultimately claimed that individual transformational rules, which included movements, were properly subsumed under general, construction-independent schema like Move q. c. It was ultimately claimed that transformations could move elements to positions higher in trees but never to lower positions. d. It was ultimately claimed that there was a system of abstract case assignment so that NPs/DPs in particular were associated with invisible cases, subject to various principles called Case Theory. e. It was ultimately claimed that there was a system of assignment (in some not clear way) of elements called q roles to DPs, subject to various principles called Theta Theory. f. It was ultimately claimed that moved phrases and their traces formed objects called chains, subject to various conditions.

250


It is not my intention to describe any of these changes in detail. They are all well known and much discussed. Accounts better than any I could provide are found in such works as Culicover (1997), Lasnik and Uriagereka (1988), Ouhalla (1994), Riemsdijk and Williams (1986), Roberts (1997), and Webelhut (1995). Roberts (1986) probably gives the description most closely tied to passive issues. These innovations and in certain cases the various theoretical losses they entail have diverse consequences for the description of passives. The recognition of traces in the two varieties null and copy, already discussed in a different context in chapter 7, need play little role in this discussion. However, not so for the shift from a view of individually formulated transformations that characterize specific constructions to one like (43b). That does have significant relations to this discussion, as seen in the early specification of it in (44): (44)

Chomsky (1986b: 72–73) “Correspondingly, the study of NP movement led to the conclusion that the various cases reduce to Move-NP. In the earliest work, there was, for example, a ‘passive transformation’ converting (29i) to (29ii) by a rule with the structural description and structural change indicated informally in (30), moving the third term to the position of the first, adding be-en to the second term see (which becomes be see-en = be seen by a later rule; we overlook here the placement of tense), to the third position where it is assigned by: (i) John saw Bill (ii) Bill was seen by John

(29)

(NP, V, NP) → (3, be-en 2, by 1)

(30)

“Similarly, the rule of raising that converts (31i) to (31ii) was expressed as a transformation (32), moving the third term of the structural description to the position of the first, which is empty in the underlying D-structure generated by phrase structure rules: (i) e seems [John to be happy] (ii) John seems [e to be happy]

(31)

(NP, V, [NP, X]) → (3, 2, 4)

(32)

“With appropriate formulation of general principles on rules and representations, both (30) and (32) reduce simply to Move-NP, so that there is no passive or raising rule but simply an interaction of principles of UG yielding various constructions, differing from language to language as a consequence of options that the languages allow. Furthermore, the differences between Move-wh, Move-NP, Move-PP, and so forth can be in large part (perhaps completely) explained in other terms, so that we are left with the rule Move-a, a being an arbitrary category. It would be too strong to claim that this conclusion has been demonstrated, but it is a reasonable hypothesis, and many particular cases appear well substantiated.”

Recourse to such schemas yields a major modification of earlier descriptions of passives like that of (3), (11), and (12). The specific formulation of passive as in, for

JUNK SYNTAX

2

251

example, (11), which simply vanishes in the ‘Move-a’ framework, in itself imposed putatively correct, although, as has been seen and will be further, massively wrong, limitations on which DPs in actives were in principle subject to passivization; it was, modulo the dodge about verb category already touched on, simply those directly postverbal DPs. But on its own, the Move-a approach imposes no conditions whatever on those DPs subject to passivization, and, as the jargon goes, vastly ‘overgenerates’. In effect then, the original transformational theory of constraints on passives, versions of which are seen in (3), (11), and (12) and (66), is abandoned. Something additional is obviously needed. An early recognition of this is seen in (45): (45)

Chomsky (1976); reprinted in Chomsky (1977a: 174) “Evidently, a grammar limited to such rules as (7) or (8) will overgenerate massively, since intricate constraints cannot be built into specific transformations. Consider the case of NP-preposing; i.e., the leftward movement case of (7). By general conditions on recoverability of deletion (the correct formulation of which is a nontrivial matter; cf. Peters and Ritchie [1973]), the second NP can move only to an NP position that is empty of any lexical material. Assuming that the left NP position, which is to receive the moved NP, is empty, either by virtue of prior NP-postposing or for some other reason, we will have such instances of NP-preposing as the following: (10)

a. b. c. d. e. f. g.

John is believed [t is incompetent] John is believed [t to be incompetent] John(‘s) was read [t book] John seems [t to like Bill] John seems [Bill to like t] yesterday was lectured t yesterday’s lecture t

“In each case, t is the trace left by movement of the NP (John, yesterday). Of these examples, only (b), (d), and (g) are grammatical, although NP-preposing has applied in a comparable way in all cases. Thus the rule overgenerates, specifically, in cases (a), (c), (e), and (f). “There are two general approaches to the problem of overgeneration in such cases as these: we may try to impose (i) conditions on the application of rules or (ii) conditions on the output of rules, i.e., on surface structures. The latter may be related to rules of semantic interpretation that determine LF, under the assumptions of EST.”

Two points about (45) are obvious.9 First, the passage does not go beyond generalities to real conditions which would properly reduce the vast excess of an unconstrained Move-a description of passivization. Second, although such descriptions no longer specify that a passivizable phrase must be immediately postverbal, the cases cited arbitrarily met this condition, which raises the question of whether the author had faced the magnitude of unwanted possibilities allowed by rule schemata. Notably, a decade later, one finds, as in (44), the reduction of passivization to a variant of the Move a-(schema being described in effect as an accomplishment, with no specification of problems or unsolved issues. What is unquestionable is that the overall issue of precisely describing the facts of passivization has largely vanished into some much vaster

252


and unfocused problem of controlling the outputs of Move-a. Given the evolutions in question, it is much less obvious that transformational claims about passives are false because it is more difficult to determine what, if anything, the relevant claims are. One additional point is that the adoption of the Move-a view abandons the proposal in (66) that compatibility with manner adverbials is a defining property of English passives, one that excludes the objects of what were called middle verbs from passivizing. One would have expected that fact to have been discussed and justified. But as far as I know, it never was. Finally, the claim of “blind application,” seen to combine prestigious apparent force with vacuity-determining hedges, of course necessarily vanishes in a Move-a(framework, implicitly granting that its only significance was the junk linguistic function of associating some unearned glamour with a factually and conceptually inadequate view.

5. The cavalier treatment of the by phrase The earliest transformational accounts of English such as (3) and (11) involved the posit of two distinct (though linked) nominal movements for passive clauses like (46a): (46)

a. Gloria was bitten by the snake. b. The snake bit Gloria.

One involved the ‘promotion’ of a VP-internal DP (here Gloria) to subject position of the auxiliary; the other, a ‘demotion’ of the underlying subject (here the snake) of an activelike structure to a position that, in some way, ended up being inside a PP. The latter mapping was problematic, as it was unclear how to guarantee the derived PP structure if, as in the earliest accounts, the structure on which passivization operated was essentially like that of the corresponding activelike (46b). One subsequent approach to this problem, that of Chomsky (1965: 104), posited an underlying P with a following empty (dummy element) NP position into which the active subject could move. A later version of this is (47): (47)

Chomsky (1976); reprinted in Chomsky (1977a: 169) “For example, we might formulate the passive transformation in English with the following SD: (3) (vbl, NP, Aux, V, NP, by, #, vbl) In this formulation, the two terms vbl are (end-) variables, so that the first and last factors of a string X to which the transformation applies are arbitrary. The second and fifth factors must be NP’s (each is an NP), the third an Aux(iliary), the fourth a V, the sixth by and the seventh # (by and # are terminal symbols; we may think of # as an ‘abstract’ representative of NP).”

In this view, the ‘demotion’ of subject part of passivization is taken to involve substitution of the subject DP for an empty post-P DP. This view was still evidently

JUNK SYNTAX

2

253

quite inadquate as, like the Syntactic Structures version, it fails to indicate that the passive by + DP form a PP. Moreover, a few years later, in Chomsky (1981), the ‘demotion’ aspect of the original transformational description of passivization had been completely abandoned. Rather, the underlying form of (46a) was claimed to be more like (48) than like (46b): (48)

[e] was bitten Gloria by the snake.

This approach establishes no obvious connection between the object of a passive by phrase and the subject of an active. One motivation for the rejection of subject postposing as a feature of passive description may have been a realization that such was incompatible with a principle like (49), already referenced in (43c): (49)

Chomsky (1975b: 107): “[T]he permissible rules are rules of ‘upgrading’ which move a noun phrase closer to the ‘root of the sentence,’ that is, to a less embedded position; the impermissible rules are rules of ‘downgrading,’ which increase the embeddedness of the noun phrase. We might stipulate, then, that upgrading rules are permitted, but not downgrading rules.”

For subject postposing in passives would, of course, be an instance of putatively banned downgrading. But other, more theory fundamental, reasons for abandoning the earlier postulation of subject ‘demotion’ are inherent in the ideas of Chomsky (1981) and subsequent work. The earlier analysis would end up violating the Theta Criterion of Chomsky (1981), as the DP moved to subject position would be assigned q roles in both that and its original position. And problems also arise from the fact that the preposed DP would apparently have to appear in a position of the trace of the postposed DP. Details are not relevant here. What is relevant, though, is that while the proposals of Chomsky (1981) abandon the demotion analysis of passive by phrases, nowhere in that work is there any explicit justification for the move or any proposed treatment of the immediate consequences of it.10 The same lack of discussion is seen in a still later work, where one is only given an indication that that the full by phrase is present at all stages of the passive: (50)

Chomsky (1986b: 73) “The same will be true of the passive rule if we assume that the structure immediately underlying the S-structure form is not (29i) but rather (33): e was see-en Bill (by John) (33)”

Inter alia, the consequences that involve partially undermining the original motivations for transformational descriptions of the passive, which included, as specified in (3), the need to capture the identical selectional/semantic relations that link active subject and passive by phrase object. There was, one recalls, even reference in (3) to the existence of NP interchange having been proved.

254


But in 1981 the author of (3) abandoned that idea without comment and with no exploration of alternative means of dealing with what had been ‘proved’ to require subject demotion. Moreover, in discussing passive it was claimed: (51)

Chomsky (1981: 124) “The traditional characterization of passive as involving a change of object to subject is correct in one important sense: this is the core case of passive.”

Not only does this fail to reference the author’s own earlier ideas, for example, (3), with their explicit recognition of a syntactic relation between active subject/ by phrase, but it also distorts the “traditional characterization,” which of course included such a relation: (52)

Jespersen (1924: 164) “In most cases this shifting is effected by means of the passive turn (B is preceded by A). Here what was the object (or one of the objects) in the active sentence is made into the subject, and what was the subject in the active sentence is expressed either by means of a propositional group, in English with by (formerly of), in French with par or de, in Latin with ab, etc., or in some languages simply by means of some case form (instrumental, ablative).”

The disdain for the development of ideas evident in the failure of (51) to mention the subject/by phrase relation, traditionally explicit as in (52), or the author’s own earlier recognition of this seen in (3), should be regarded as another junk linguistic feature. If there is no transformational relation between active subject/passive by phrase object, then the unchallenged selectional/semantic relations that link them (see, e.g., Baker, Johnson and Roberts, 1989; Jaeggli, 1986: 599; and Roberts, 1986: 27) must be described by some mechanisms M distinct from transformational derivation. Chomsky (1981, 1986b) provided no account of M or even any indication that it was necessary. This issue does not seem to have been faced in transformational terms until Jaeggli (1986) and Baker (1988a: chapter 6). These works proposed idiosyncratic accounts of mechanism M. Both accounts appealed to manipulations of so-called q roles, arguably the most obscure and undeveloped aspect of the framework at issue.11 For Jaeggli (1986: 600) the basic idea was “that the passive suffix en is crucially involved in transferring the external q-role onto the NP in the by-phrase in a passive sentence.” This was elaborated in a complex way. Baker (1988a: 335–336) refined and expanded Jaeggli’s proposal by claiming that the external q role of the verb is assigned to the passive morpheme and that the by phrase ‘doubles’ the q role of the passive morpheme. The latter is taken to parallel the features of pronominal clitics. To partly formalize these ideas, Baker proposed a special type of coindexing, which linked the passive morpheme and the by phrase. It seems fair to see in these complexities a forced, far-fetched attempt to recapture part of what the simple identity of active subject and passive by phrase recognized in other approaches, which include that of (3), elegantly yielded. Moreover, the question arises why mechanisms of the type M, however elaborated, could not supply an alternative to that part of the DP movement description of English passives, DP preposing, which post-1981 transformational accounts main-

JUNK SYNTAX

2

255

tained. Surely, one reply to this question would reference idiom pieces, for one of the standard arguments for a transformational description of English passives has for decades appealed to idiomatic pieces of VPs, as in (53); see, for example, Culicover (1976: 167–168) and Radford (1988: 422–424). (53)

a. The doctors took/*brought/*found/*grabbed/*grasped little heed of her problems. b. Little heed was taken/*brought/*found/*grabbed/*grasped of her problems by the doctors.

The claim has been that the transformational aspect of the passive is revealed not merely by the selectional/semantic relations but precisely by the passivizability of pieces of structure that either are not directly semantic or do not bear a direct semantic relation to the verb in whose clause they passivize. Exactly this view was expressed by Chomsky (1975b: 114), where it was claimed that if there were no passives such as inter alia those that involved idioms, there would be no motivation for a transformational derivation of passives. The idea is that the relations can be stated only once with respect to the active structure, with the transformational movement as it were projecting them to passives without cost. This argument about idioms is of course maintained even under the altered, ‘demotion’-rejecting transformational conception of passives; see, for example, Chomsky (1981: 85). From this point of view then, rejection of the earlier transformational derivation of the passive by phrase object claims that there should not be corresponding subject demotion cases. That is, the post-1981 transformational view of the passive implicitly denies that there are instances in which an idiomatic subject piece of a structure T appears as the by phrase object of a passive that corresponds to T. Such an absence would apparently justify a difference between capturing the object/subject selectional similarities via one mechanism (e.g., movement) but the subject/by phrase ones by another, some instantiation of M. Testing this claim, evidently, requires first finding active clauses that have both idiomatic subjects and otherwise passivizable objects. However, if the claims of Marantz (1984: 26–27) were fully correct, there could be none. Marantz discussed a putative asymmetry in which there are many verb + object idioms that combine with nonidiomatic subjects but apparently no subject + verb idioms that combine with nonidiomatic, nonfixed objects. While Marantz noted the existence of idioms like (54a), he stressed that in these the object is as fixed as the subject; in any event, no passive is possible, as in (54b, c): (54)

a. The shit hit the fan. (‘things went bad’) b. *The fan was hit. (ok a physical fan was struck; *‘things went bad’) c. The fan was hit by the shit (*‘things went bad’)

However, while English cases of the sort Marantz denied are not common, they are, contrary to his claims, not nonexistent, as illustrated first in (55): (55)

a. The lovebug bit Ted/your uncle/the guy next door/several policemen last year = “Ted/etc. fell in love last year” b. The lovebug has bitten Ted/your uncle/the guy next door/several policemen again.

256


There should be no doubt about idiomaticity here; the term lovebug occurs only with the verb bite, and this verb in such contexts does not denote dental activities. (56)

The lovebug *chewed/*clawed/*grabbed/*licked/*mangled/*overwhelmed/*seized Ted.

Expressions like (55) have two properties that Marantz claimed could not exist: (i) an idiomatic subject coexisting with a freely choosable (human) object and (ii) an idiomatic subject occurring in expressions whose overall form is not that of a fixed phrase. So the post-1981 ‘no demotion’ transformational view of English passivization determines that cases like (55) have no corresponding grammatical passives where the idiomatic subject phrase of the active occurs as object of a passive by phrase But this is false; they do:12 (57)

a. Ted was bitten by the lovebug. b. Most romantics are bitten by the lovebug at least once during their lives.

Moreover, the same morphological elements bite and bug also form a distinct but more productive idiomatic sequence. This is illustrated by (58) from Bresnan (2001: 15), credited to Avery Andrews: (58)

The photography bug has bitten/*chewed up/*disturbed/*killed Fred.

That is, there seems to be an idiom of the form [DP D + Nounx + bug] Aux bite DP2; the whole is interpreted to mean something like “DP2 has become strongly interested in Noun.” While the verb seems to be fixed as bite and bug must be present, Nounx is seemingly not fixed as long as it can be taken to designate an activity of people; that is, roughly Nounx must be able to appear also in the context [DP is into [DP (D) Nounx + (plural)]: (59)

a. The chess/golf/heroin/Internet/poker/polo/skiing/television/tennis/video game/ categorial grammar bug has bitten Fred. b. Fred is into chess/heroin/the Internet/poker/polo/skiing/swimming/television/video games/categorial grammar.

As in the idiomatic usage in (55), passivization is possible: (60)

Fred has been bitten by the chess/golf/heroin/Internet/poker/polo/skiing/television/ tennis/video game/categorial grammar bug.

A third case of the same sort as (55) is, I think, seen in (61): (61)

a. A little bird told/*promised/*sang/*wrote/*informed me (of) that. b. I was told/*promised/*sung/*informed (of) that by a little bird.

The stars in (61b) mark only the case where a little bird is interpreted idiomatically as designating not a small object of possible ornithological interest but a person/source that the speaker is explicitly indicating (s)he is refraining from identifying.

JUNK SYNTAX

2

257

Further, as mentioned in note 6 of chapter 2, Nunberg, Sag, and Wasow (1994) observed that certain figurative idiom pieces like birds of a feather can occur in restricted cases as subjects of certain control complements. Now, a few control structures like that of (62) permit a passive whose subject is extraposition it: (62)

a. Those kids decided to hang out together. b. It was decided by those kids to hang out together.

Notably then, (63b) seems to me no worse than (62b), while (63d) has the impossible character of (63c): (63)

a. b. c. d.

Birds of a feather may decide to flock together. It may be decided by birds of a feather to flock together. *Birds of a feather may decide to surf the Net. *It may be decided by birds of a feather to surf the Net.

This amounts then to a fourth case where an idiomatic active subject piece can appear in a passive by phrase. And a fifth is provided by (64), where only the noncanine reading of old dogs is relevant: (64)

a. b. c. d.

Old dogs may even decide to learn new tricks. It may even be decided by old dogs to learn new tricks. *Old dogs may even decide to stop drinking. *It may even be decided by old dogs to stop drinking.

Contrary to expectations under a view where the passive by phrase is not syntactically related to active subjects, clear cases have been found of subject idiom chunks in passive by phrases. Since for transformational views the passivizability of idiomatic VP phrase pieces is taken to support a DP ‘promotion’ of passive subject/ active object relations, parity of reasoning indicates that the facts in (57), (60), (61b), (63b), and (64) support a ‘demotional’ view of passive by phrase/active subject relations.13 This does not yield support for a transformational treatment of passivization; it just means that within that framework an asymmetric ‘promotion only’ view is factually untenable. It also falls out that the claims of Baker (1988a) and Jaeggli (1986) that the relations between active clause subject DPs and passive by phrase DPs can be taken as completely mediated by the manipulation of q-roles is untenable. Even if there are such things, they cannot suffice for this function. To conclude, the history of the treatment of passive by phrases in transformational terms is an instructive illustration of junk linguistics. Beginning with the exaggerated claims in (3) of a proof that the object DPs of passive by had to be moved subjects, one finds an evolution where the ‘proved’ view is later not even taken as true. And when it was abandoned, no reference to the earlier claims was made and a distorted view of pregenerative views of passive was offered. Moreover, the abandonment was linked to no explicit discussion of how to deal with the facts that motivated what had been claimed to have been proved. Subsequent work like that of

258


Baker (1988a) and Jaeggli (1986) grappled openly with the problems created but offered only rather convoluted appeals to obscure q-role notions. And the evolved view yields proposals that inherently make the false claim that passive by phrase objects cannot be the sort of restricted DPs whose behavior as passive subjects has standardly been taken to justify transformational movement. Moreover, no facts appear to conflict with the view that by phrase objects are realizations of syntactic subjects. The historical rejection of this idea seems to have been entirely conceptdriven. So, nearly a half-century after the claimed provably needed DP interchange in (3), there evolved a set of ideas arguably descriptively worse with respect to passive by phrases than the original 1957 view.14

6. Reanalysis Previous to the conceptual evolutions listed in (43), the transformational view in (3), (11), and (12) faced, as alluded to in section 2, a vast problem of undergeneration with respect to pseudopassives like (65b): (65)

a. Myron referred to the problem. b. The problem was referred to by Myron.

Such cases seem to fall outside the schema of description in (3), (11), and (12) since the passivized phrase is apparently not immediately postverbal in what would be the passive rule input. There seem to have been two different transformational approaches to this issue, a transient one in Chomsky (1965) and a standard one. The former was described as follows: (66)

Chomsky (1965: 103–104) “These observations suggest that the Manner Adverbial should have as one of its realizations a ‘dummy element’ signifing that the passive transformation must obligatorily apply. That is, we may have the rule (55) as a rewriting rule of the base and may formulate the passive transformation so as to apply to strings of the form (56), with an elementary transformation that substitutes the first NP for the dummy element passive and places the second NP in the position of the first NP: (55) Manner → by ∩ passive (56)

NP— Aux—— V—. . .—NP —. . .-by- passive -

(where the leftmost . . . in (56) requires further specification—e.g., it cannot contain an NP). “This formulation has several advantages over that presented in earlier work on transformational grammar (such as Chomsky, 1957). First of all, it accounts automatically for the restriction of passivization to Verbs that take Manner Adverbials freely. That is, a Verb will appear in the frame (56) and thus undergo the passive transformation only if it is positively specified, in the lexicon, for the strict subcategorization feature [—NP ∩ Manner], in which case it will also take Manner Adverbials freely.

JUNK SYNTAX

2

259

Second, with this formulation it is possible to account for the derived Phrase-marker of the passive by the rules for substitution transformations. This makes it possible to dispense entirely with an ad hoc rule of derived constituent structure that, in fact, was motivated solely by the passive construction (cf. Chomsky, 1957, pp. 73–74). Third, it is now possible to account for ‘pseudopassives,’ such as ‘the proposal was vehemently argued against,’ ‘the new course of action was agreed on,’ ‘John is looked up to by everyone,’ by a slight generalization of the ordinary passive transformation. In fact, the schema (56) already permits these passives. Thus ‘everyone looks up to John by passive’ meets the condition (56), with John as the second NP, and it will be converted into ‘John is looked up to by everyone’ by the same elementary transformation that forms ‘John was seen by everyone’ from ‘everyone saw John.’”

The key currently relevant idea in this proposal is the variable represented by the first set of three dots, constrained not to contain an NP. This broadens the class of NPs/DPs allowed to prepose in passives from the immediately postverbal ones picked out by (3). It basically allows anything to occur between the verb and a passivizable DP except a DP. Even though (66) would properly allow passivization of postverbal PP objects, it hardly represented a correct analysis of these. First, it was still too restricted; even the vast freedom of choice would not have permitted description of, for example, (14b). Moreover, the only principle (66) offered distinguishing V + PP structures that yielded pseudopassives from those that did not was compatibility with manner adverbs. It is seen in the following discussion of reanalysis that this is nowhere near correct. Moreover, it would arguably have wrongly allowed passivization of DPs out of infinitives, as in (67): (67)

a. Jerome decided in an obsessed way. b. *Europe was decided to become famous in in an obsessed way by Jerome.

In any event, (66) seemingly vanished from the transformational framework without a trace (no pun intended), and without any arguments being offered for abandoning it. Instead, the standard resolution for pseudopassives came to appeal to something called reanalysis, a device more often invoked than defined. So Pesetsky (1995: 275) rightly says: “The preposition in these constructions must be affixed to V by a morphological process whose exact nature is unclear.” The core assumption is that an operation like (68a) can remove a preposition from a PP and incorporate it into a preceding verb, converting a Verb + PP structure into one of the form Verb + DP, hence subject to the transformational passive defined on postverbal DPs.15 This idea originates, as (68b) shows, in pregenerative work, although Curme never seems to be credited: (68)

a. Reanalysis (schematic)

Skeptical Linguistic Essays

Skeptical linguistic essays

Skeptical linguistic essays